Rules of exponence
Following Zwicky (1985), it is conventional to divide realization rules into two basic types: rules of exponence, which spell features out by forms, and rules of referral, which relate the realization of distinct feature spell-outs. The rules in Figure 6.2, illustrate the flexibility of exponence rules. As the partitive rule illustrates, a realizational model can express a biunique correlation between the interpretation of a single feature, here partitive case, and the realization of a minimal form unit, here -a. Yet the partitive case feature remains associated with the full word forms taloa and taloja, not with the exponent -a alone. Moreover, the one-to-one relation between the case feature and the case marker that ‘spells it out’ is not normative, but merely a limiting case.
Because exponence rules interpret features rather than ‘cancel’ or ‘consume’ them, they can describe feature-form mismatches in both directions: ‘fusional’ patterns in which there are more features than forms, and ‘fissional’ patterns in which there are more forms than features. The nominative and genitive plural rules in Figure 6.2 illustrate a fusional pattern, which Matthews (1972: 67) terms cumulative exponence. In this example, it is multiple features (case and number) that are ‘cumulatively’ associated with single form units. The converse mismatch, which Matthews (1972:93) terms extended exponence, is not exhibited by a single exponence rule but by a set of exponence rules that realize partially overlapping feature bundles. Extended exponence is exhibited by the Latin form re:ksisti: ‘you had ruled’ in Figure 2.3 and by the Ancient Greek form elelykete ‘you had unfastened, in Figures 3.1 and 3.2. The rules that Matthews (1991:177) proposes to describe elelykete are summarised below.
Figure 6.4 Root form and 2nd plural active paradigm cell of Greek lyo ‘unfasten’
Figure 6.5 Perfective ‘shortening’ rule
Figure 6.6 Perfective reduplication rule
The analysis starts from the structure in Figure 6.4, which combines the features of the ‘2nd person plural past perfective indicative active’ paradigm cell with the root form ly: of the verb lyo ‘unfasten’. This structure does not represent the root entry of lyo, which would only contain whatever intrinsic features are taken to characterize uninflected roots, but instead contains the features of the paradigm cell that is realized by the surface word form elelykete. This surface form is obtained by applying exponence rules that interpret the features in Figure 6.4 by successively modifying the root form.
To illustrate extended exponence, it will suffice to review the feature overlap in the rules that define the perfective active stem lelyk. The shortening rule in Figure 6.5 applies first. This rule realizes perfective aspect by reducing a long root vowel, shortening ly: to ly. The reduplication rule in Figure 6.6 then repeats the initial stem consonant, obtaining lely from ly. The rule in Figure 6.7 applies next, defining the perfective active stem lelyk by suffixing -k to lely.
Figure 6.8 summarizes the ordering of exponence rules, the overlapping features that they realize, and the cumulative effects of their application. Different rule orders would yield the same output in this case, provided that the shortening rule precedes the suffixation of -k. But it is only the effect on the shape of the input form that constrains the order of rule application. Since exponence rules
Figure 6.7 Active perfective stem rule
Figure 6.8 Spell-out of the active perfective stem lelyk
are purely interpretive, the features associated with the paradigm cell in Figure 6.4 remain constant through the analysis, irrespective of the order in which rules are applied. It is the fact that spell-out does not alter features that allows each of the rules in figures 6.5-67 to realize the same perfective feature. Vowel shortening, reduplication and suffixation of -k all have the same status as exponents of the feature perfective, and there is no obvious sense in which any one marker is ‘primary’ and the others are ‘secondary’.
In sum, the feature-form separation permits a degree of flexibility in the way that features and forms are related by exponence rules. Distinctive features may be biuniquely realized, as in Post-Bloomfieldian models. A morphemic pattern is illustrated by the rule in Figure 6.2 that realizes plural features by -i. But multiple features may also be realized cumulatively, as illustrated by the rule in Figure 6.2 that realizes nominative and plural features by -t. Features may also be present but remain unrealized, as in the case of nominative singular features, which define a cell in Figure 6.3 but are not realized by any rule. Conversely, features may be realized, individually or cumulatively, by multiple rules. This pattern is illustrated by the three rules that realize perfective in Figure 6.8. None of these patterns are unusual or rare, and even more extreme cases of extended exponence are discussed in Harris (2009).
-  Matthews (1991) subscripts the category label ‘V’ to identify word class, rather than including itamong the features to be realized, though nothing hinges on whether category labels are included withinfeature bundles or subscripted to them. To maintain consistency with Matthews’s formulations, thephonemic forms in these rules are not placed between slashes. Thus Figure 6.5 relates a form ending ina long vowel, /XV:/, to a form /XV/, which is identical except that it ends in the short counterpart /V/.
-  Accent placement is suppressed, as it is predictable here (Goodwin 1894:29).