Within a system, multiple patterns of rule combination may interact in complex ways. I present a detailed analysis of Swahili verb inflection in which simple rules, composite rules and aggregated rules all enter into intricate competition, yielding an extravaganza of deviations from canonical morphotactic criteria. At the center of this discussion are three characteristics of Swahili conjugation: (i) the polyfunctionality of verbal concords (in virtue of which the same rule is used to express the noun class of a verb’s subject, that of its pronominal object, or that of a relativized argument), (ii) the expression of negation (by means of three complementary rules), and (iii) the marking of relative verb forms (whose relativized-argument affix participates in an extensive pattern of affix counterposition). The rule-combining approach to morphotactics allows the interacting details of these subsystems to be resolved into two very general types exhibiting an unexpected degree of economy.

I discuss the general implications of the rule-combining approach to morphotactics developed in the course of the foregoing chapters. I summarize the numerous superficially problematic phenomena that the rule-combining approach resolves and I relate these phenomena to the variety of ways in which rule combinations may deviate from the canonical characteristics of a language’s morphotactics. Finally, I synopsize the set of formal definitions on which the rule-combining approach is based.

Besides affording a way of modeling deviations from canonical morphotactics, rule composition makes it possible to see apparently recalcitrant morphotactic patterns as conforming to canonical criteria if these are assumed to cover composite rules as well as simple rules. I examine an apparent deviation from the integral stem criterion in Sanskrit and apparent deviations from the rule opposition criterion in Latin, Limbu, and Sanskrit. Each of these phenomena can be reconciled with the canonical criterion from which it apparently deviates if this criterion is assumed to cover composite rules as well as simple rules. All of these are cases in which deviation from the minimal rule criterion facilitates conformity to other canonical morphotactic criteria.

I discuss my assumptions about individual morphological rules, and then explain the ways in which rules combine. Rule composition, the default mode of rule combination, models many canonical patterns, but also models certain deviations from the canonical morphotactic criteria. Holistic rule combination, a second mode of combination, accounts for deviations from the compositional content criterion, in which a rule combination realizes more than the sum of the content that those rules realize individually. Rule aggregation, a third mode of combination, accounts for deviations from the stem operand criterion, in which a rule operates not on a stem, but on the affix introduced by the rule with which it combines. Counterpotentiation, a fourth mode of combination, accounts for deviations from the intermediate well-formedness criterion, in which the result of applying one rule is ill-formed unless its application is followed by that of another particular rule. I outline the elaboration of these ideas in the ensuing chapters.

Rule composition makes it possible to model an important kind of deviation from the rule independence criterion – cases in which the application of one rule is directly dependent on that of another, carrier rule. In such cases, the only use of the dependent rule is as part of a composite rule incorporating both the dependent and its carrier. Where the definition of a word form involves two carrier rules, it can further happen that the same dependent rule composes with both of them, engendering a pattern of multiple exponence that deviates from the unique sequence criterion. I discuss two cases of this sort of rule dependency: Limbu verb morphology exhibits a pattern in which dependent rules compose with their carrier rules; Sanskrit presents a pattern of the reverse sort, in which a carrier rule composes with its dependent.

I discuss different kinds of deviation from the parallel sequence criterion; I illustrate with detailed examples from Fula, Udmurt, and Eastern Mari. In Fula verb inflection, rules of subject and object marking involve a default applicational sequence that is overridden in specific circumstances by the opposite sequence of application; this deviation can by modeled by postulating two patterns of rule composition, one realizing the default sequence and the other overriding that default. Udmurt noun inflection is different, since it involves two patterns of rule composition that do not stand in a default/override relation but are instead simply complementary. Nevertheless, the Fula evidence and the Udmurt evidence both conform to the unique sequence criterion. The declensional morphology of Eastern Mari, by contrast, deviates from that criterion, since it allows alternative acceptable sequences of rule application; in the rule-combining approach to morphotactics, these can be seen as involving alternative patterns of rule composition realizing the same morphosyntactic content.

Languages often present cases in which two rules together express content some of which cannot be attributed to either rule individually. The inflection of regular verbs in Breton presents several examples of this sort; holistic rule combinations are a way of modeling this phenomenon. Some instances of holistic combination have an emergent character; these are cases in which all of the forms realized by the composite of two rules happen to possess a property P that neither rule realizes on its own. In such cases, the composite is open to reanalysis as a holistic combination realizing P. Limbu verb morphology provides an example of this sort. A particularly compelling case for the postulation of holistic combinations arises in systems in which the same two rules express different holistic content in different contexts; Old English verb inflection is a system of this sort.

A familiar phenomenon in derivational morphology is potentiation – the creation of new contexts for some affix by the prior addition of some other affix; thus, ‑able potentiates ‑ity in English. Less familiar is the reverse phenomenon of counterpotentiation – the licensing of an affix by the subsequent addition of some other affix. In English, -al counterpotentiates -ic in forms such as whimsical, which has no counterpart in -ic without -al (*whimsic). Facilitatory relations of these two sorts reflect distinct modes of rule combination. The potentiation of an outer rule by an inner rule is a kind of rule composition in which the domain-of-definition of the composite is that of the inner rule; the counterpotentiation of an inner rule by an outer rule is a mode of rule combination in which the domain-of-definition of the combination is disjoint from that of the inner rule. Potentiation enhances conformity to the intermediate well‑formedness criterion; counterpotentiation is a deviation from this criterion.

Affix counterposition has two subcases: in the first (exemplified by the inflection of reflexive verbs in Lithuanian), an affix that is suffixed to the stem in some words is suffixed to a prefix in others; in the second, mirror-image case (exemplified by Noon adjective concord), an affix that is prefixed to the stem in some words is prefixed to a suffix in others. Rule aggregation models the phenomenon of affix counterposition as a deviation from the stem operand criterion involving a rule of affixation R whose operand is ordinarily a stem (when R isn’t aggregated) but is instead an affix (when R is aggregated to the rule introducing that affix). Rule aggregation brings affix counterposition into conformity with the affix directionality criterion. Nevertheless, there are real deviations from the latter criterion: some languages have true ambifixes that actually function as prefixes in some word forms but as suffixes in others. Gurma noun-class inflection exemplifies this possibility. Moreover, the morphotactics of Italian pronominal affixes involves a significant interaction between true ambifixation and rule aggregation.

In the rule-combining approach to morphotactics, the same rules may compose in more than one way to express more than one content. Strikingly, Murrinhpatha verb inflection allows the same rules to compose in the same linear order but in different binary groupings to express distinct content. As a consequence of this fact, Murrinhpatha verb forms exhibit a systematic pattern of ambiguity, as illustrated by pubamngankungkardungime (loosely, ‘they saw us’), in which the paucal nonsibling female suffix -ngime may relate to the verb’s object (allowing the interpretation ‘those two siblings saw us (paucal nonsibling females)’) or to its subject (allowing the interpretation ‘they (paucal nonsibling females) saw us two siblings’). As I show, this ambiguity follows from the Category Determination Principle, according to which a rule whose morphosyntactic content is ambiguous is disambiguated by the first rule with which it composes. I give a detailed demonstration that Murrinhpatha verb inflection exploits this fact.

Situating the rule-combining approach to morphotactics in a wider theoretical context, I summarize its implications for the architecture of Paradigm Function Morphology and for schema unification in Construction Morphology. I further contrast the exponence-driven conception of morphotactics embodied by the rule-combining approach with the very different word-skeletal approach to morphotactics, drawing attention to two advantages of the rule-combining approach. First, it affords a more parsimonious inventory of morphological operations than is assumed in Distributed Morphology. Second, it avoids the cumbersome theoretical commitments of Information-based Morphology (the assumptions of position-based ordering, rule anchoring, and distributional pigeonholing), which entail numerous complications in the analysis of a language’s morphotactics. The distinct assumptions on which the rule-combining approach rests (those of combination-based ordering, unanchored rules, and distributional multidetermination) afford morphotactic analyses that are at once simpler and more explanatory.

In this chapter, I discuss the preliminary assumptions of the rule-combining approach to morphotactics and advance the two fundamental hypotheses that underlie it: the morphotactic holism hypothesis and the morphotactic variety hypothesis (Section 1.2). In Section 1.3, I review previous proposals that provide empirical support for the morphotactic holism hypothesis, which (unlike the morphotactic variety hypothesis) is not a novel idea. In Section 1.4, I discuss the nature of canonical morphotactics, for which I introduce ten criterial characteristics, construed in rule-based terms. In Section 1.5, I give examples of phenomena that possess these characteristics as well as of phenomena that do not apparently possess them. The morphotactic phenomena to be analyzed in the following chapters deviate from some of these canonical characteristics, but reinforce conformity to others provided that a rule-combining approach is assumed. In Section 1.6, I anticipate the range of topics to be discussed in subsequent chapters.

Rule combination can contribute to morphological simplicity. Synchronically, rule combinations (like word combinations) are sometimes stored as formulaic units, and this fact contributes to a morphological system’s processing simplicity, since accessing a stored rule combination directly is simpler than decomposing that combination into its component rules for separate lookup. Stored, formulaic rule combinations may also contribute to diachronic simplifications of a language’s morphology, since they are the locus of reanalyses that may eventuate in “affix telescoping,” the development of a rule combination into a simple rule. But affix telescoping is not a monolithic phenomenon; it involves the reduction of a rule combination’s combinatory transparency along at least four dimensions. Thus, it is possible to find rule combinations that are progressing toward reanalysis as simple rules without yet having reached the point of reanalysis.