The 'item and pattern' model
The designations ‘IA’ and ‘IP’ both classify frameworks in terms of combinatoric strategies (involving arrangements or processes) and leave models to specify the types of units that are concatenated or modified. In contrast, Hockett (1954) does not characterize a ‘WP’ in terms of general combinatorial strategies, but in terms of a pair of specific units: words and (inflectional) paradigms. Hence the ‘WP model’ described by Hockett and Robins is indeed a model in the sense proposed above, instantiating a more abstract framework. This underlying framework is based on implicational relations, often between properties and forms or between pairings of properties and forms. The most accurate designation for this framework in Hockett’s terms would be item and pattern. In realizational models that instantiate this framework, the distinctive patterns are defined by the features in abstract paradigm cells (or property bundles) that trigger realization rules. Items correspond in turn to the output of those rules. More conservative implicational models formalize counterparts of the exemplary paradigms and principal parts of a classical grammar. In this variant, it is possible to identify patterns by reference to paradigms whose cells contain properties and forms, items consist of a single pairing of properties and forms, and the comparison of items against patterns may be used to sanction the deduction of items via processes of analogical extension or, more generally, on the basis of correlations. A distinctive property of this second approach is that it defines implicational structure over networks of interrelated
‘units as wholes’, and abstracts smaller units from these networks. In treating parts as ‘abstractions’ over larger wholes, the resulting neo-Saussurean conception inverts the Bloomfieldian view that wholes are ‘constructed’ from smaller parts.
From this perspective, WP models, both realizational and implicational, can be seen as instantiations of an item and pattern framework that is ‘tuned’ to the structure of a flexional language. A highly salient morphological feature of these languages is the grammatical role and predictive value of words and paradigms. The privileged status of words reflects the fact that, as Robins (1959:128) puts it, “the word is a more stable and solid focus of grammatical relations than the component morpheme by itself”. Hence it is often possible to associate determinate properties with words, even when they are composed of grammatically or semantically indeterminate parts. The role of paradigms derives likewise from the fact that the feature space of an inflectional paradigm is essentially closed and uniform. This allows a realizational model to define the abstract cells of an inflectional system independently of the forms that realize those cells. The closed and uniform structure of inflectional paradigms also provides a maximally reliable analogical base for deducing new forms based on previously encountered forms. Taken together, the informativeness of words and the interdependence of cells in a paradigm determine “the... general insight... that one inflection tends to predict another” (Matthews 1991:197).
Recognizing WP models as specific instantiations of a more general implica- tional framework helps to clarify how item and pattern approaches can be applied to languages in which words and paradigms play a less significant role. Consider first isolating patterns and derivational formations. In an isolating language, standard inflectional paradigms will not guide deductions about novel forms. Yet other sets of word forms may still establish patterns that are of predictive value. ‘Morphological families’ consisting of sets of derivationally-related forms exhibit their own patterns of interpredictability, which in some cases are as reliable as the expectations generated by inflectional paradigms. Both in size and composition, these ‘families’ show far more item-specific variation than inflectional paradigms. Whereas inflectional paradigms are broadly uniform within a given word or inflection class, families of derivational formations can vary in size by orders of magnitude. The ‘lexical neighbours’ of an item may also be of deductive value, as may be other word classes. The idea that an inflectional paradigm is the extreme case of a predictive pattern is implicit in the way that the notion ‘paradigm’ is extended to broader classes of related forms in Robins (1959:121) and Moscoso del Prado Martin (2003).
Identifying implicational relations as the cornerstone of WP models also avoids the need to impose a uniform analysis on all languages at the level of units. Although it is reasonably well established that most languages do not conform to the ‘agglutinative ideal’ of a morpheme-based model, there is no reason in principle why some languages could not approach this ideal.7 In a perfectly morphemic language, grammatical properties would reliably predict the formatives that realize them (modulo regular phonological processes), and formatives would reliably signal the properties that they realize. In such a language, implication would retain a central role, but shift its locus from relations between words to relations between formatives and grammatical properties.
An implicational model also integrates the patterns that Aronoff (1994) classes as ‘morphomic’. Much of the contemporary discussion of‘morphomes’ focusses on the descriptive problems that they pose for analyses that attempt to assign them a determinate analysis in isolation. From an implicational perspective, it is not the patterns themselves that are problematic but the assumption that the components of the patterns can be analyzed separately. Implicit in classical and philological treatments of morphomic patterns is the view that they serve a predictive or diagnostic function (Maiden 2005). This relational view is most clearly expressed in the description of the patterns that Matthews (1972) terms ‘Priscianic’ or ‘parasitic’. In these patterns, a form in one part of an inflectional paradigm sanctions reliable predictions about the shape of another. The fact that systematic correspondences that are orthogonal to the distinctive grammatical properties of the synchronic system maybe maintained—and extended (Maiden 2005)—calls into question the role of shared properties in more ‘normative’ cases of exponence. From an impli- cational standpoint, it is the ‘structuralist morpheme’ rather than the ‘Priscianic morphome’ that is the special case, as this pattern abstracts away from the larger systems to which a form belongs and establishes a fixed relation between properties and forms strictly at the level of the formative.8 Hence whereas morphomes do not fit well into a morphemic account, an account that treats morphomes as units of predictive value subsumes morphemes as morphomes that encapsulate a biunique prediction between properties and forms.
In sum, the pivotal role of words and paradigms in WP approaches reflects the high information load that words carry within the implicational networks defined by paradigms and other morphological ‘families’. The stable grammatical information associated with a word not only serves to identify its own function paradigms were treated as sets of word-categorization pairs; from Lieb (1983) lexical words are treated as paradigm-meaning pairs, with ‘meaning’ a psychological concept. By allowing ‘improper paradigms’, this approach extends a paradigmatic treatment to arbitrary parts of speech in arbitrary languages, including Chinese (Sackman 2000; Su 2011). A paradigmatic treatment is not extended to derivational families which are treated as ‘word families’ or sets of lexical words (Lieb 2013).
- 7 See Hockett (1987: §7.4) for discussion of the origins of the agglutinative ideal.
- 8 Morphological patterns of this type can be described solely by ‘rules of exponence’ (Zwicky 1985) which spell out the features of a single paradigm cell or feature bundle ‘locally’ by units of form, without the need for ‘rules of referral’ or other strategies that relate the realization of multiple forms. However, as discussed in Chapters 5-7, there is little empirical evidence that morphological uncertainty can in general be resolved effectively at the formative level, and considerable experimental evidence that speakers do not in fact process individual words in isolation from other words.
but also locates it within an inflectional paradigm and within the larger morphological system. In this way, the grammatical information associated with a form facilitates deductions about other forms, based on systematic correspondences and interdependencies within a language. Aspects of form that sanction implications likewise express a type of information, information which can be modelled by notions developed within information theory (Shannon 1948). In the case of an inflectional paradigm, the informativeness of a form correlates with the degree to which knowledge of that form reduces uncertainty about other forms in the same paradigm. The notion of uncertainty reduction is implicit in the use of ‘diagnostic’ principal parts in classical accounts, though from a more explicit information- theoretic perspective, nearly all forms are informative to some degree about others.
An item and pattern model is thus well-adapted to exploiting the stable information content of word-sized units and the tight implicational structure determined by inflectional paradigms in languages where these units and patterns are both present. However, these units and structures are best regarded not as parts of the model per se but as recurrent elements in language that facilitate the application of an item and pattern model. Maintaining a clear distinction between frameworks and their instantiations offers a useful perspective on the intrinsic properties of item and pattern models as well as on the scope and limits of WP approaches, classical as well as modern, that instantiate them.
-  These families have well-attested effects on language processing, and it is hard to imagine that theseeffects do not in some way reflect the organization of a speaker’s lexicon. See, among others, Schreuderand Baayen (1997), de Jong (2002), Kostic et al. (2003), Moscoso del Prado Martin et al. (2004b), andthe discussion in Chapter 7.
-  The fact that members of a given word or inflection class tend to have a comparable number offorms helps to account for the observation that the processing of inflectional forms is more sensitiveto token frequency whereas the processing of derivational formations is more sensitive to the typefrequency of related items (Baayen et al. 1997).
-  See Pounder (2000) for a similar extension of ‘paradigm’ to derivational families. Models ofintegrational linguistics (fn. 2 above) propose more general paradigmatic extensions. Until Lieb (1980),