The first stage in the rehabilitation of the WP model was essentially complete by the reprinting of Robins (1959) in 2001, nearly a half century after the plea for equal treatment in Hockett (1954). The realizational approaches developed during this time were, as Hockett had anticipated, “clearly distinct from either IA or IP” and had decisively corrected “the erroneous impression that there were principally just two archetypes” (386). But other issues remained unresolved. One issue concerned how faithfully realizational models represented the “traditional framework for the discussion of Latin, Greek, Sanskrit” that Hockett had originally had in mind. Another concerned the applicability of realizational WP models to isolating or agglutinative patterns or, indeed, to derivational processes. These issues had been set aside to some degree while the viability of WP approaches was being established in the inflectional domain, where classical models had been most successfully applied. It was only by reexamining the model underlying these approaches that the rejuvenated WP tradition could begin to assess the suitability of WP as a general model of morphological analysis, and to address the types of objections that had led proponents of IA and IP models initially to neglect WP approaches.
It is worth recalling that Hockett’s initial endorsement of the WP model had not been unqualified. Instead, it had carried the caveat “that it is obviously insufficiently general, incapable of organizing efficiently the facts of a language like Chinese”, voicing one of the primary concerns about the model. The response offered by Robins (1959) sets the tone for much of the subsequent literature. Rather than addressing the objection directly, Robins deflects it by shifting the terms of debate onto the question of whether WP analyses of isolating languages might be correct in identifying words as minimal units:
To say that paradigms are not part of a model applied to a language whose words do not exhibit grammatical paradigms is labouring the obvious, but it may still be urged that the word as formally established is the most profitable unit to be taken as basic in the statement of the sentence structures of such languages... (Robins 1959:123)
This retreat to an essentially word-based perspective reflects Robins’ general view that the descriptive success of the WP model derives principally from the treatment of words as minimally meaningful units. However, Robins appears to regard even the status of words as negotiable, and leaves open the possibility of assigning morphemic analyses to genuinely agglutinative languages:
The segmentation into morphemes need not take into account any need for a parallel representation of the grammatical categories applicable in every word in any class (though, of course, it may do so if a clear statement on these lines is possible). (Robins 1959:133, emphasis added)
Behind this theoretical pragmatism lay a deeper agnosticism. Early proponents of the modern WP model began with the modest goal of establishing the model on the same footing as IA and IP models. Having achieved that goal, they saw no point in replacing overblown universalist claims for IA or IP models with similar claims for the WP model. Robins (1959) and Matthews (1972), in particular, were openly skeptical about whether any single model of morphology could be applied with equal success to all types of languages:
It may also be that while each of the models discussed in this paper is feasible with every language, one of them is more appropriate with certain languages; possibly Mixteco, at least on Pike’s analysis, is not a ‘WP language’, and certainly on the evidence we have considered some languages are less suitably ‘IA’ or ‘IP languages’. (Robins 1959:144)
In particular, there is no reason to assume (pace Hockett, 1954, and others) that the same model of description must be equally applicable to all languages. The opposite view may be more illuminating. (Matthews 1965:141)
Finally, it has become clear at least that different languages raise quite different problems in morphological analysis. It is therefore possible that they also require quite different sorts of description. (Matthews 1972:156)
Hockett’s original admission that “we have no completely adequate model” had reflected dissatisfaction with technical solutions to problems of analysis created by IA and IP models, especially when applied to languages of the flectional type. The caution expressed by Robins and Matthews acknowledged the complementary obstacles that the WP model faced in analyzing the types of languages that IA and IP models had been expressly designed to describe. The most immediate empirical challenges lay in showing how the WP model could avoid imposing inflectional paradigms on isolating languages like Chinese, or avoid treating words as basic in agglutinating languages like Turkish.
To a large extent, these challenges grew out of the strategy of retrofitting a formal model onto an established descriptive tradition. The problem is present from the outset of the modern revival when Hockett (1954) characterizes ancient grammatical descriptions as embodying a ‘word and paradigm’ model. This designation accurately emphasizes the role that words and inflectional paradigms playin classical WP approaches. However, it begs the question whether these components define the WP model, or merely specify the units and structures to which the model has been most fruitfully applied.
Although this question arises in a particularly acute form for the WP model, it reflects a more general source of unclarity in Hockett’s classification. The detailed elaboration of IA and IP approaches in Hockett (1954) never explicitly distinguishes between general frameworks of analysis and the models that instantiate frameworks by specifying substantive assumptions, including assumptions about units of analysis. The IA framework that emerges from Hockett’s description is fundamentally concatenative in nature, representing word structure by means of combinations of sub-word units. The models that instantiate IA frameworks are typically ‘atomistic’ in that they operate with segmentally minimal units. However, these models could just as well concatenate larger units, such as stems, or even sequences of affixes that frequently collocate and pattern like units. The defining property is that word structure (and word meaning) arises through the concatenation of units.
The central innovation in the IP framework is the introduction of processes that apply operations other than concatenation to a base. Unlike realization rules, processes do not ‘spell out’ previously specified features. Rather they are featuremodifying or ‘incremental’ in the sense of Stump (2001). Just as the concatenation of a morpheme to a base yields a unit that augments the features and form of the base, the application of a process may alter the features and modify the form associated with an input to which it applies. However, models that instantiate IP frameworks are again free to specify different types of inputs (roots, stems or other units), along with different formal operations.