Within a single system of morphotactics, multiple patterns of rule combination may interact in complex ways. In this chapter, I demonstrate this fact by presenting a detailed analysis of a significant fragment of Swahili verb inflection in which simple rules, composite rules, and aggregated rules all enter into competition.Footnote ¹ As the examples in Table 8.1 show, the Swahili system of verb inflection presents an extravaganza of deviations (some real, others merely apparent) from a number of canonical morphotactic criteria, including the rule independence criterion, the parallel sequence criterion, the unique sequence criterion, the stem operand criterion, the rule opposition criterion, and the affix directionality criterion.

The rule‑combining approach to morphotactics, though it itself inherently deviates from the minimal rule criterion (Section 1.5.1), allows deviations from the last two of these criteria to be seen as merely apparent; moreover, it affords an explanatory way of modeling genuine deviations from the other canonical criteria. I shall argue that despite the morphotactic complexity of the Swahili conjugational system, the rule-combining approach to morphotactics makes it possible to model this system in a very simple way at the most general level. As I shall show, the rules of Swahili verb inflection fall into three groups (Groups III, II, and I) such that the definition of a verb form’s morphotactics a composite of either the type (R_III ◦ R_II) or the type ((R_III ◦ R_II) ◦ R_I). The real complexity of the analysis resides in identifying the group membership of the system’s simple rules and their many and varied combinations.

At the center of this discussion are three characteristics of the Swahili conjugational system: the polyfunctionality of verbal concords (Section 8.2), the expression of negation (Section 8.3), and the special marking of verb forms for a relativized argument (Section 8.4). I begin with a brief description of the wider morphosyntactic setting of these three characteristics.

8.1 The Morphotactics of Swahili Verb Inflection

The broader structural context in which these characteristics exist are no doubt familiar to many readers. As a Bantu language, Swahili possesses an array of noun classes whose effects pervade its entire inflectional system. The Swahili noun classes may be distinguished as in Table 8.2.Footnote ² In this table, a count noun’s gender is represented as a pairing ncx+ncy of noun classes such that the noun’s singular and plural forms belong to ncx and ncy, respectively. Because noncount nouns don’t distinguish singular and plural numbers, their gender is therefore an unpaired noun class (nc14).Footnote ³

As Table 8.2 shows, each noun class is associated with a prefix. In the simplest cases, a noun exhibits the prefix associated with the noun class to which it belongs, as the examples in the table show. (The exact form of a noun-class prefix may naturally vary according to its phonological context; thus, the noun class 1 prefix is m‑ in m‑toto ‘child’, mw‑ in mw‑ongo ‘liar’, and mu‑ in mu‑umbaji ‘creator’.) There are, however, cases in which a noun simply seems to exhibit the wrong prefixes; for example, the singular noun kipofu ‘blind man’ (plural vipofu) belongs to noun class nc1 (plural nc2) despite exhibiting ki- (plural vi-) rather than m- (plural wa-) as its prefix. Thus, a noun’s prefix isn’t always a reliable sign of its noun-class membership.

Definitive evidence of a noun’s noun-class membership is provided by an expression that agrees with it by means of a concordial affix. Thus, in the sentence Kipofu aliondoka ‘A blind man left’, the concordial subject prefix a- on the verb form aliondoka ‘(s/he) left’ reveals that kipofu belongs to nc1. In some cases, the concordial affixes take the same form as the corresponding noun‑class prefix; thus, in Watu watalala ‘people will sleep’, the noun-class prefix of wa-tu ‘people’ (a member of nc2) is reflected by the nc2 subject concord wa- of the verb form wa-talala ‘(they) will sleep’. Often, however, the verbal concords for a given noun class differ from the noun-class prefix associated with that class; thus, in Mtu atalala ‘a person will sleep’, the noun-class prefix of m-tu ‘person’ (a member of nc1) differs from the corresponding nc1 subject concord a- of the verb form a-talala ‘(s/he) will sleep’. The inventory of Swahili verbal concords corresponding to the noun classes in Table 8.2 is listed in Table 8.3; these verbal concords include first‑ and second‑person prefixes (singular and plural) as well as third‑person prefixes distinguished according to gender as well as number.

8.1.2 The Morphotactics of Negation in Swahili Verb Inflection

The problems posed by verbal negation in Swahili are twofold. First, negation isn’t always expressed in the same place in a verb form’s morphotactics, contrary to the parallel sequence criterion. Second, the exponence of negation is complex: in one class of cases, negation is expressed cumulatively with subject agreement; in another class of cases, it is expressed cumulatively with tense; and in still other cases, its expression is cumulative with neither subject agreement nor tense. As a consequence, negative morphology gives rise to apparent deviations from the rule opposition criterion and to genuine instances of multiple exponence.

The negation of an indicative verb form is ordinarily effected by a rule ⟦ha‑⟧ that introduces the prefix ha‑; thus, alongside the positive form tutasoma ‘we will read’ in (1a) is the negative form hatutasoma ‘we will not read’ in (4a). This general pattern leads one to expect that the first‑person singular form nitasoma ‘I will read’ in (1b) should have *hanitasoma as its negative counterpart, but it doesn’t; instead, the corresponding negative form is the form sitasoma ‘I will not read’, as in (4b), in which the expected sequence of the negative prefix ha‑ and the first‑person singular subject concord ni‑ is supplanted by the portmanteau prefix si‑.Footnote ⁴

(4)

a.	ha‑tu‑ta‑soma	b.	si‑ta‑soma
	neg‑sbj.1pl‑fut‑read		neg.1sg‑fut‑read
	‘we will not read’		‘I will not read’

The morphotactics of negation is further complicated in two ways. First, there is a second negative prefix having the form si‑ that differs both paradigmatically and syntagmatically from the portmanteau prefix si‑. Unlike the portmanteau prefix si‑, which expresses both negation and first‑person singular subject agreement in indicative forms such as (4b), this second negative si‑ doesn’t express subject agreement, but appears in nonindicative contexts in the presence of any sort of subject, as for example in the prohibitive subjunctive form in (5b). The nonindicative negative si‑ also differs syntagmatically from the portmanteau negative si‑: the former appears alongside a separate subject-agreement prefix (as in (5c)) while the latter never does (cf. (5d)), and the latter appears alongside a separate tense prefix (as in (4b)) while the former never does (cf. (5e)). That is, the portmanteau si‑ is paradigmatically opposed to the subject‑agreement prefixes while the nonindicative si‑ is paradigmatically opposed to the tense prefixes, as in Figure 8.1.

(5)

a.	u‑ni‑saidi-e
	sbj.2sg‑obj.1sg‑help-sbjv
	‘help me’
b.	u‑si‑ni‑saidi-e
	sbj.2sg‑neg.nonind‑obj.1sg‑help-sbjv
	‘don’t help me’
c.	ili	ni‑si‑som-e
	so.that	sbj.1sg‑neg.nonind‑read-sbjv
	‘so that I don’t read’
d.	*ni‑si‑ta‑soma
	sbj.1sg‑neg.1sg‑fut‑read
	Purportedly: ‘I will not read’
e.	*ili	ni‑si‑ta‑som-e
	so.that	sbj.1sg‑neg.nonind‑fut‑read-sbjv
	Purportedly: ‘so that I will not read’

Figure 8.1 Two distinct patterns of paradigmatic opposition

A second additional complication in the morphotactics of negative indicative verb forms is the incidence of multiple exponence in negative past‑tense forms. In each of the negative verb forms considered so far, negation has only a single exponent: the prefix ha‑ in (4a), the portmanteau prefix si‑ in (4b), and the nonindicative prefix si‑ in (5b,c). In the past tense, however, negation involves the negative past‑tense prefix ku‑ in addition to either ha‑ (as in (6b)) or si‑ (as in (7b)). These forms raise natural questions. What induces the application of ⟦ha‑⟧ in hatukusoma ‘we did not read’ insofar as the application of ⟦ku‑⟧ itself realizes negation? Why isn’t *nikusoma admissible as a nonredundant alternative to the multiple exponence of negation in sikusoma ‘I did not read’? I return to these questions in Section 8.3, where I show how principles of rule combination make it possible to give a precise account of the morphotactics of negation in Swahili verb inflection.

(6)

a.	tu‑li‑soma	b.	ha‑tu‑ku‑soma
	1pl‑pst‑read		neg‑1pl‑neg.pst‑read
	‘we read’		‘we did not read’

(7)

a.	ni‑li‑soma	b.	si‑ku‑soma
	1sg‑pst‑read		neg.1sg‑neg.pst‑read
	‘I read’		‘I did not read’

8.2 The Subject/Object Polyfunctionality of Verbal Concords in Swahili Verb Morphology: Fragment I

In the analyses that I propose here and in Sections 8.3–8.4, rules of Swahili verb inflection fall into three groups; in Figure 8.2, the affixes of Swahili verb inflection are classified according to the rule group by which they are defined. All groups contain both simple rules and combined rules. The numbering of the three groups reflects the order in which a (simple or combined) member of one group combines with a (simple or combined) member of another group. Group III includes the rule ⟦ha‑⟧ introducing the indicative negative prefix ha‑, the rules introducing the subject concord prefixes, and combinations of these rules. Group II includes the rules introducing the tense prefixes, the rule introducing the nonindicative negative prefix si‑, the rules introducing the relative affixes, and combinations of these rules. Group I includes the rules introducing the pronominal object prefixes.

Figure 8.2 Individual affixes of Swahili verb inflection categorized according to the rule group by which they are defined

The morphosyntactic property sets realized by the rules in Groups I–III are somewhat complex in their structure. Besides inflecting for mood, tense, and polarity, a Swahili verb inflects for subject agreement (indicating the subject’s person and number, and in the third person, its noun class); a verb may also inflect for the person, number, and noun class of a pronominal object, and as we have seen, relative verb forms further inflect for the number and noun class of a relativized argument. In the preliminary Fragment I of Swahili verb inflection under scrutiny in this section, the morphosyntactic property sets of positive indicative verb forms include the properties in the ‘Fragment I’ row of Table 8.5. This system of property sets is extended in Sections 8.3–8.4 to additionally accommodate the properties associated with subjunctive, negative, and relative verb forms.

The morphosyntactic properties in Table 8.5 do not combine absolutely freely; instead, a well‑formed morphosyntactic property set must conform to various property coöccurrence restrictions. For Fragment I, the relevant restrictions are those in (11). The notion ‘extension’ to which (11d–e) refer is defined recursively in (12); the ‘atomic properties’ in this analysis are those listed in Table 8.5.

For the preliminary Fragment I of Swahili verb inflection under discussion in this section, the basic members of the three groups of rules of exponence are those in (13). For expository convenience, I employ three conventions in the representation of these rules. First, each rule label ⟦x⟧ carries an external, Roman‑numeral subscript indicating its group membership; for example, the Group III rule introducing the negative prefix ha‑ has the label ⟦ha‑⟧_III. Second, if a rule ⟦x⟧ introduces an affix x that expresses noun class n, I represent x with a bold subscript n; for instance, the Group III rule introducing the subject‑agreement prefix a‑ for noun class 1 is labeled ⟦a‑₁⟧_III. Third, I use the ⟦id.fcn⟧ label for rules that realize specific content through the application of an identity function; there are two such rules in (13), distinguished by their rule‑group subscript: ⟦id.fcn⟧_I, ⟦id.fcn⟧_III.

In Fragment I, the basic rules of exponence in Group III express subject agreement; the basic rules in Group II express tense; and the basic rules in Group III express a pronominal object.

Not included among the basic rules in (13) are the “versatile” rules that introduce affixes that may serve both to express subject agreement and to express the properties of a pronominal object. The rules in (14) are all versatile in this way. The first‑person singular ⟦ni‑⟧ rule, for example, applies to express a first‑person singular subject in the definition of forms such as nitasoma ‘I will read’ but also applies to express a first‑person singular object in forms such as atanipiga ‘s/he will strike me’.

The versatility of the rules in (14) follows from the fact that these rules realize category-indeterminate properties (Section 2.1.1). For example, the property set {{1 sg}} realized by the application of ⟦ni-⟧ contains the set-based property {1 sg}, which is indeterminate with respect to its inflectional category. In the definition of some forms, ⟦ni-⟧ participates in the realization of the set-based property {sbj 1 sg} (which belongs to the inflectional category of subject agreement); in the definition of other forms, it participates in the realization of the set-based property {obj 1 sg} (which belongs to the inflectional category of pronominal object specification). In accordance with the Category Determination Principle (Section 2.1.1),Footnote ⁷ ⟦ni‑⟧ always combines with a rule realizing a category-determinate property Q such that the unification [{1 sg} ⊔ Q] is itself a set-based property that is well-formed and category-determinate; the same is true of all of the other rules in (14).

Whether a versatile rule of this sort expresses subject agreement or the properties of a pronominal object thus depends on how it combines with other rules. To account for this fact, I assume that the rules in (14) are dependent rules that do not inherently belong to any of the three rule groups but may compose with the carrier rules given in (13) for Groups I and III,Footnote ⁸ that is, the rules ⟦id.fcn⟧_I and ⟦id.fcn⟧_III.Footnote ⁹ Their combination with these carrier rules conforms to the pair of patterns in (15).Footnote ¹⁰

The composite rules defined for Group III by (15a) are listed in (16); those defined for Group I by (15b) are listed in (17). Whether the dependent rules in (14) express properties of subject agreement or properties of a pronominal object depends on their carrier rule: when one of the dependent rules in (14) composes with the carrier rule ⟦id.fcn⟧_III in (13a), its content unifies with the carrier rule’s content {{sbj}}, producing a composite expressing subject agreement, as in (16); when that same dependent rule composes with the carrier rule ⟦id.fcn⟧_I in (13c), its content unifies with the carrier rule’s content {{obj}}, producing a composite expressing the properties of a pronominal object, as in (17).Footnote ¹¹

Given this approach, the full exponence rules (FERs) for Fragment I can be characterized through the composition of rules from Groups I, II, and III, as in (18). (As will be seen, the two patterns in (18) extend without modification to Fragments II and III.)

Examples of the FERs defined by (18) are the composite rules in (19). The FER (19a) defines forms such as ninasoma ‘I am reading’, as in part (a) of Table 8.6; here, the composite rule (⟦ni‑⟧ ◦ ⟦id.fcn⟧_III) in (16a) expresses subject agreement. The FER (19b) defines forms such as atampiga ‘s/he will strike her/him’, as in part (b) of Table 8.6; in this form, the subject properties and the pronominal object properties are respectively realized by the simple rules ⟦a‑₁⟧_III and ⟦m‑₁⟧_I in (13a,c). The FER (19c) defines forms such as watawapiga ‘they will strike them’, as in part (c) of Table 8.6; here, the dependent rule ⟦wa‑₂⟧ in (14c) figures in both the expression of subject agreement (by means of the composite rule (⟦wa‑₂⟧ ◦ ⟦id.fcn⟧_III) in (16c)) and the expression of a pronominal object (by means of the composite rule (⟦wa‑₂⟧ ◦ ⟦id.fcn⟧_I) in (17c)).

This way of accounting for the subject/object polyfunctionality of the rules in (14) correctly portrays these rules as deviations from the unique sequence criterion: each of the rules in (14) may join with other rules in more than one sequence, owing to the fact that it is a dependent rule whose carrier may belong either to Group III or to Group I.Footnote ¹² The fact that the rules in (14) may express either properties of subject agreement or properties of a pronominal object is not attributed to their own formulation, but to the formulation of the carrier rule with which they compose: in accordance with the Category Determination Principle, combination with the carrier rule ⟦id.fcn⟧_III in (13a) produces a composite rule expressing subject agreement, while combination with the carrier rule ⟦id.fcn⟧_I in (13c) produces a composite rule expressing the properties of a pronominal object.

8.3 The Morphotactics of Negation in Swahili Verb Morphology: Fragment II

Building on the foregoing analysis, I now examine a larger fragment of Swahili conjugational morphotactics. Fragment II includes negative forms and (nonindicative) subjunctive forms, so the morphosyntactic properties ‘–ind’, ‘neg’, and ‘sbjv’ are now included in the relevant inventory of morphosyntactic properties. (See the ‘Fragment II’ row of Table 8.5.) With the inclusion of these additional properties, the property coöccurrence restrictions in (11) must now be supplemented with the restrictions in (20). According to (20a), subjunctive forms are necessarily nonindicative; according to (20b), indicative forms are all tensed forms (though, as we will see in Section 8.4, the reverse is not always true).

In order to account for negative and subjunctive forms, it is necessary to add some rules of exponence to Groups III and II; these are the additional rules in (21).

The basic rules of exponence in Fragment II are the result of combining the basic rules for Fragment I (i.e. the rules in (13)) with the basic rules in (21). Taken together, the basic members of Group III in (13a) and (21a) are quite varied. They include rules expressing indicative negation (⟦ha‑⟧_III), subject agreement (⟦u‑⟧_III, ⟦m‑⟧_III, ⟦a‑₁⟧_III), or both (⟦si‑⟧_III). The basic members of Group II in (13b) and (21b) are similarly varied, since they include rules expressing tense (⟦ta‑⟧_II, ⟦li‑⟧_II, ⟦na‑⟧_II), nonindicative negation (⟦si‑⟧_II), tense cumulatively with (indicative) negation (⟦ku‑⟧_II and ⟦-i⟧_II), and subjunctive mood (⟦-e⟧_II). The ⟦-i⟧_II and ⟦-e⟧_II rules in (21b) differ from the other basic members of Group II in that they introduce suffixes. I assume that these suffixes -i and -e induce the elision of a stem-final vowel (as in ha-tu-tak-i ‘we do not want’ [stem -taka], u‑ni‑saidi-e ‘help me!’ [stem -saidia]).Footnote ¹³ In addition to its basic members, Group II includes the composite rule in (22), which serves in the inflection of negative subjunctive forms such as u‑si‑som-e ‘do not read!’.

Group III also includes composite rules – those in (16) and some new ones. The latter arise through the composition of the indicative negative ⟦ha‑⟧_III rule with the basic subject‑agreement rules in (13a) (i.e. with ⟦u‑⟧_III, ⟦m‑⟧_III, and ⟦a‑₁⟧_III) and with the Group III composites in (16). Thus, the composite rules in Group III for Fragment II are those of Fragment I plus those in (23). Note that ⟦ha‑⟧_III does not compose with ⟦ni‑⟧_III; the portmanteau rule ⟦si‑⟧_III in (21a) takes the place of this composite.

Exactly as in Fragment I, the FERs for Fragment II are characterized by the FER patterns in (18). Examples of the FERs that (18) defines for Fragment II are the composite rules in (24). The FER (24a) defines forms such as hatutasoma ‘we will not read’, as in part (a) of Table 8.7. Here, the Group III rule is the composite (23a) and the Group II rule is the future‑tense rule ⟦ta‑⟧_II; because hatutasoma lacks a pronominal object, no Group I rule applies in this FER. The FER (24b) defines forms such as hatukumpiga ‘we did not strike her/him’ as in part (b) of Table 8.7, accounting for the multiple exponence of negation by means of the negative composite (23a) in Group III and the negative past‑tense rule ⟦ku‑⟧_II in Group II. The FER (24c) defines forms such as sikumpiga ‘I didn’t strike her/him’ by means of a composite of three simple rules: the first‑person singular negative portmanteau rule ⟦si‑⟧_III in Group III, the negative past‑tense rule ⟦ku‑⟧_II in Group II, and the third‑person singular (noun class 1) rule ⟦m‑₁⟧_I in Group I; see part (c) of Table 8.7. The FERs (24d,e) define subjunctive forms such as unisaidie ‘help me!’ (in which ⟦-e⟧_II serves as the Group II rule) and usinisaidie ‘don’t help me!’ (which the composite negative subjunctive rule (⟦si‑⟧_II ◦ ⟦-e⟧_II) in (22) serves as the Group II rule); see parts (d) and (e) of Table 8.7.

This analysis accounts for two ways in which the morphotactics of Swahili negative verb forms apparently deviates from canonical criteria. First, it accounts for the fact that in Swahili, negative morphology does not conform to the parallel sequence criterion, according to which rules that express the same inflectional category canonically occupy the same position in any sequence of rule applications. In Swahili, negation is expressed sometimes by a Group III rule (as in hatutasoma ‘we won’t read’, sitasoma ‘I won’t read’), sometimes by a Group II rule (as in usinisaidie ‘don’t help me’), and sometimes by both a Group III rule and a Group II rule (as in hatukusoma ‘we didn’t read’, sikusoma ‘I didn’t read’).

Second, the analysis proposed here accounts for an apparent deviation from the rule opposition criterion, according to which relations of paradigmatic opposition among rules of inflectional exponence are, canonically, relations among individual rules. In Swahili, the portmanteau ⟦si‑⟧_III rule introducing the first‑person singular negative prefix si‑ in forms like sitasoma ‘I won’t read’ apparently stands in paradigmatic opposition to the sequence of rules introducing the prefix string ha‑m‑ in second‑person plural negative forms like hamtasoma ‘you (pl.) won’t read’; but if the latter rules are assumed to constitute a composite rule (⟦ha‑⟧_III ◦ ⟦m‑⟧_III), then the relation of paradigmatic opposition between ⟦si‑⟧_III and (⟦ha‑⟧_III ◦ ⟦m‑⟧_III) actually conforms to the rule opposition criterion.Footnote ¹⁴

The proposed analysis of negation in Swahili conjugation also accounts for issues relating to multiple exponence (Section 8.1.2). Consider first the form hatukusoma ‘we didn’t read’. What necessitates the appearance of the negative prefix ha‑ in this form given the presence of the negative past‑tense prefix ku‑? In the analysis proposed here, hatukusoma arises from the form cell (25), whose realization, in accordance with (18a), involves the narrowest applicable rules from Groups II and III. (Note that there is no rule in Group I that is applicable for the realization of (25).) The composite rule (⟦ha‑⟧_III ◦ (⟦tu‑⟧ ◦ ⟦id.fcn⟧_III)) (= (23a)) overrides (⟦tu‑⟧ ◦ ⟦id.fcn⟧_III) (= (16b)) as the narrowest applicable rule in Group III. Because the past‑tense ⟦li‑⟧_II rule in (13b) isn’t as narrowly applicable as the negative past‑tense ⟦ku‑⟧_II rule in (21b), the latter is the only Group II rule applicable for the realization of (25). This fact has no effect on the choice of rule in Group III. That is, what necessitates the appearance of the negative prefix ha‑ in hatukusoma is the fact that (in accordance with (18a)) the FER for (25) is a two‑rule composite (R_III ◦ R_II) such that each of R_III and R_II is the narrowest applicable rule in its group.

Now consider the form sikusoma ‘I didn’t read’, in which negation is expressed cumulatively with first‑person singular subject agreement by the prefix si‑ and cumulatively with past tense by the prefix ku‑. What excludes the possibility of the nonredundant form *ni‑ku‑soma (in which first‑person singular subject agreement is expressed by ni‑, as in positive forms) rather than the redundant form si‑ku‑soma as a realization for the form cell in (26)? The answer is that the portmanteau ⟦si‑⟧_III rule in (21a) overrides the composite rule (⟦ni‑⟧_III ◦ ⟦id.fcn⟧_III) in (16a) as the narrowest rule in Group III applicable for the realization of this cell, in accordance with (18a). As with (25), the negative past‑tense ⟦ku‑⟧_II rule is the only Group II rule applicable for the realization of (26).

8.4 The Morphotactics of Relativization in Swahili Verb Morphology: Fragment III

I now extend the proposed fragment of Swahili verb inflection to include relative verb forms. In order to accommodate these in the analysis proposed in Sections 8.2–8.3, several additions are necessary. First, I assume the additional morphosyntactic properties ‘–tense’ and ‘rel’ in the ‘Fragment III’ row of Table 8.5; the property ‘–tense’ is associated with tenseless relative verb forms and the property ‘rel’ is associated with all relative verb forms. Combinations of the properties in Table 8.5 are subject to the revised set of property coöccurrence restrictions in (27), which subsume and extend those given above in (11) and (20). By restrictions (27e) and (27f), subjunctive verb forms and relative verb forms are both nonindicative; relative verb forms are seen as nonindicative insofar as they are never used on their own to perform the speech act of assertion in the way that indicative verb forms are. By restriction (27g), forms in this fragment that are specified as tenseless are relative, but by (27h), only negative relatives are necessarily specified as tenseless; that is, positive relatives may or may not express overt tense. Restrictions (27j,k) entail that just as the properties ‘sbj’ and ‘obj’ identify nested property sets encoding the person, number, and gender of a verb form’s subject and pronominal object arguments, the property ‘rel’ identifies nested property sets encoding the person, number, and gender of a relative verb form’s relativized argument. Restriction (27l) guarantees that the argument coded by a verb’s relative morphology will likewise be coded by either its subject or its object morphology.

In Fragment III, three additional basic rules of exponence are necessary; these are all members of Group II, and are formulated in (28). The ⟦taka‑⟧_II rule expresses the future tense in relative verb forms. The basic rules encoding a relative verb form’s relative argument are ⟦‑ye₁⟧_II and ⟦‑o⟧_II. The ⟦‑ye₁⟧_II rule encodes a relativized argument belonging to noun class 1 through the suffixation of ‑ye; by contrast, ⟦‑o⟧_II is a default rule that encodes the presence of a relativized argument without expressing any particular noun class.

In this extended fragment, I propose to treat relativized-argument rules (whether simple or combined) as rules of suffixation that participate in the definition of affix counterpositions in the presence of a tense/negation prefix; thus, the suffixational ⟦‑ye₁⟧_II rule in (28) is responsible both for the -ye suffix in the tenseless form (mtu) asomaye ‘(person) who reads’ and for the -ye suffix that is counterposed to the tense prefix taka- in (mtu) atakayesoma ‘(person) who will read’. In order to accommodate this analysis, Group II includes a large number of aggregated rules. These are of two sorts: aggregated relativized-argument rules and aggregated relative tense/negation rules. Consider these in turn.

The dependent argument-coding rules that realize nonpersonal noun classes (= (14d–m)) aggregate with the default relativized‑argument rule ⟦‑o⟧_II to produce aggregated relativized‑argument rules, in accordance with the pattern in (29); like their carrier rule ⟦‑o⟧_II, these aggregated rules belong to Group II. The aggregated rules defined by (29) are listed in (30). For expository convenience, I abbreviate each of the aggregated rules in (30) as in the column to its right; these abbreviations highlight the fact that the rule aggregations in (30) define suffixational counterpositions. These aggregated rules are another instance of the versatility of the noun-class rules (14d–m). Each rule R in (14d–m) realizes a category-indeterminate property P; but in accordance with the Category Determination Principle, the corresponding aggregation (R Ⓐ ⟦‑o⟧_II) in (30) realizes the well-formed, category-determinate property [P ⊔ {rel}].

In accordance with the pattern in (31), the relativized argument rules in Group II (including the basic rules ⟦‑ye₁⟧_II and ⟦‑o⟧_II in (28) as well as the aggregated rules in (30)) aggregate with Group II rules of tense and nonindicative negation to produce aggregated relative tense/negation rules; these are listed in Table 8.8. Like their component rules, all of the aggregated rules in Table 8.8 belong to Group II. Unlike the aggregations in (30) (which define suffixational counterpositions), the aggregations in Table 8.8 define prefixational counterpositions.

It is worth taking a moment to notice how heterogeneous Group II has become. It includes

• simple rules of prefixation (e.g. the simple tense rules in (13b), (21b), and (28) and the simple rule ⟦si‑⟧_II of nonindicative negation in (21b));

• simple rules of suffixation (e.g. the negative definite time rule ⟦-i⟧_II and subjunctive rule ⟦-e⟧_II in (21b) and the simple relativized-argument rules ⟦‑ye₁⟧_II and ⟦‑o⟧_II in (28));

• the composite negative subjunctive rule (⟦si-⟧_II ◦ ⟦-e⟧_II) in (22), which involves both prefixation and suffixation;

• the aggregated relativized-argument rules in (30), which are suffixational; and

• the aggregated relative tense/negation rules in Table 8.8, which are prefixational.

All of these rules are in paradigmatic opposition, and a single principle – Pāṇini’s principle – suffices to resolve all instances of competition among them.

A couple of additional features of the proposed account of relative verb forms should be carefully noted. First, the ⟦ta‑⟧_II rule in (13b) is not applicable in the realization of relative verb forms because it realizes indicative mood as well as future tense; recall again that relative verb forms are nonindicative. Second, the basic relativized argument rule ⟦‑o ⟧_II in (28) is, again, a default rule that doesn’t express any particular noun class, but it does apply by default in the realization of a plural relativized argument of noun class 2; this is because neither the relativized‑argument rules in (28) nor the aggregated relativized-argument rules in (30) include a dedicated rule for noun class 2. The aggregations of ⟦‑o⟧_II in Table 8.8 (namely (⟦‑o⟧_II Ⓐ ⟦taka‑⟧_II), (⟦‑o⟧_II Ⓐ ⟦na‑⟧_II), (⟦‑o⟧_II Ⓐ ⟦li‑⟧_II), and (⟦‑o⟧_II Ⓐ ⟦si‑⟧_II)) are likewise default rules that do not express any noun class; as such, they are usually overridden by the other aggregations in Table 8.8, but they, too, apply by default in the expression of plural relativized arguments belonging to noun class 2, for which no dedicated aggregations exist in Table 8.8.

As before, the FERs for Fragment III are defined through the composition of rules from Groups I, II, and III, in accordance with the FER patterns in (18). Examples of the FERs that (18) defines for Fragment III are the composite rules in (32). The FER in (32a) defines tenseless positive verb forms such as asomaye ‘(person) who reads’, as in part (a) of Table 8.9. The fact that the verb form is unspecified for either tense or negation allows ⟦‑ye₁⟧_II, the basic relativized‑argument rule for noun class 1, to operate directly on the verb stem soma; encoding the same person, number, and gender as the simple subject‑agreement rule ⟦a‑₁⟧_III, ⟦‑ye₁⟧_II here expresses a relativized subject.

The FER in (32b) defines tenseless positive forms such as wasomao ‘(people) who read’, as in part (b) of Table 8.9. In this form, the composite subject‑agreement rule (⟦wa‑₂⟧ ◦ ⟦id.fcn⟧_III) expresses noun class 2; because there is no dedicated relativized‑argument rule for noun class 2, the default relativized‑argument rule ⟦‑o⟧_II instead applies, operating directly on the verb stem soma to express a relativized subject.

The FER in (32c) defines tenseless positive forms such as tuvisomavyo ‘(books) which we read’, in which the aggregated relativized‑argument rule ⟦‑vyo₈⟧_II matches the person, number, and gender of the composite pronominal-object rule (⟦vi‑₈⟧ ◦ ⟦id.fcn⟧_I) and therefore expresses a relativized object, as in part (c) of Table 8.9.

The FER in (32d) defines positive past‑tense forms such as tulivyovisoma ‘(books) which we read’. Here, the aggregated relativized‑argument rule ⟦‑vyo₈⟧_II (which again expresses a relativized object) is itself aggregated with the past-tense rule ⟦li‑⟧_II to produce the aggregated relative tense rule (⟦‑vyo₈⟧_II Ⓐ ⟦li‑⟧_II), as in part (d) of Table 8.9.

The FER in (32e) defines tenseless negative forms such as tusivyovisoma ‘(books) which we do not read’. In this instance, ⟦‑vyo₈⟧_II (once again expressing a relativized object) is aggregated with the nonindicative negative rule ⟦si‑⟧_II to produce the negative relativized‑argument rule (⟦‑vyo₈⟧_II Ⓐ ⟦si‑⟧_II), as in part (e) of Table 8.9.

8.5 Discussion and Conclusions

The analysis of the Swahili conjugational system presented cumulatively in Fragments I, II, and III draws attention to the fact that Swahili verbs are highly uniform in their morphotactic characteristics. In accordance with the FER patterns in (18), all three fragments consist of finite forms whose FERs are either of the type (R_III ◦ R_II) or of the type ((R_III ◦ R_II) ◦ R_I), where R_I, R_II, and R_III are rules belonging to Groups I, II, and III, respectively. As Table 8.10 shows, this is the case whether a verb form is indicative, subjunctive, or relative and whether it is positive or negative.

Table 8.10 FERs are of either the type (R_III ◦ R_II) or the type ((R_III ◦ R_II) ◦ R_I) in the morphotactics of Swahili verbs

R_I, R_II, and R_III are rules belonging to Groups I, II, and III, respectively.

R_SBJ is a (simple or composite) subject‑agreement rule.

R_TNS is a simple tense rule.

R_OBJ is a (simple or composite) pronominal object rule.

R_REL is a (simple or aggregated) relativized‑argument rule.

	Positive forms			Negative forms
	RIII	RII	RI	RIII	RII	RI
Indicative	RSBJ	RTNS	(ROBJ)	$\left\{\begin{array}{l}\left({〚\mathbf{ha}〛}_{\mathrm{III}}\circ {\mathrm{R}}_{\mathrm{SBJ}}\right)\\ {〚\mathbf{si}〛}_{\mathrm{III}}\end{array}\right\}$	RTNS	(ROBJ)
Subjunctive	RSBJ	⟦-e⟧II	(ROBJ)	RSBJ	(⟦si-⟧II ◦ ⟦-e⟧II)	(ROBJ)
Relative	RSBJ	$\left\{\begin{array}{l}{\text{R}}_{\text{REL}}\\ \left({\text{R}}_{\text{REL}}Ⓐ{\text{R}}_{\text{TNS}}\right)\end{array}\right\}$	(ROBJ)	RSBJ	(RREL Ⓐ ⟦si‑⟧II)	(ROBJ)

The proposed analysis elucidates the ways in which this system deviates from certain canonical morphotactic criteria but also reveals the ways in which, contrary to first impressions, it actually conforms to some criteria.

According to the rule independence criterion, a rule’s operation on a stem is conditioned only by that stem’s characteristics, and hence is not directly sensitive to the concurrent application of some other rule. In the inflection of Swahili verbs, however, there are many dependent/carrier relations: the argument‑coding rules in (14) only apply in combination with one of three carrier rules (the Group III carrier rule ⟦id.fcn⟧_III in (13a), the Group II carrier rule ⟦‑o⟧_II in (28), or the Group I carrier rule ⟦id.fcn⟧_I in (13c)); the specific property set realized by each of the rules in (14) depends entirely on the unification of its content with that of a carrier rule.

We have already seen how the proposed analysis models the deviation from the unique sequence criterion in the morphotactics of argument coding (Section 8.2); cf. again the fact that the ⟦wa-₂⟧ rule applies both before and after the future-tense rule ⟦ta-⟧_II in the analysis of watawapiga ‘they will strike them’ in part (c) of Table 8.6. We have likewise seen how the proposed analysis models the deviation from the parallel sequence criterion in the morphotactics of negation (Section 8.3); cf. again the fact that negation may be expressed by a Group III rule, by a Group II rule, or by both the Group III and the Group II rule in certain composites, as in the analyses in Table 8.7.

According to the stem operand criterion, the formal operation associated with a morphological rule is an operation on stems. Each of the aggregations in (30) and Table 8.8 is a case in which a rule deviates from this criterion, operating on an affix rather than on a stem. The Swahili facts are doubly interesting in this regard insofar as most of the rules in Table 8.8 involve nested aggregations – the suffixation of a prefixed suffix to a prefix, for example ‑livyo: ((⟦vi‑₈⟧ Ⓐ ⟦‑o⟧_II) Ⓐ ⟦li‑⟧_II).

The Swahili relativized‑argument rules seem to introduce affixes that are sometimes prefixal and sometimes suffixal, contrary to the affix directionality criterion. The argument‑coding rule ⟦vi‑₈⟧ in (14i) introduces a prefix in the definition of tu‑ta‑vi‑soma ‘we will read them’ but seems also to introduce a suffix in the definition of the tenseless relative form tuvisomavyo (← tu‑vi‑soma‑vi‑o) ‘(books) that we read’; the relativized‑argument rule ⟦‑ye₁⟧_II in (28) introduces a suffix in the definition of the tenseless relative form a‑some‑ye ‘(person) who reads’ but seems to introduce a prefix in a‑taka‑ye‑soma ‘(person) who will read’. Although the aggregation operation inherently deviates from the stem operand criterion, it also allows us to maintain that the dependent argument‑coding rules in (14) are uniformly prefixational and that the simple and aggregated relativized‑argument rules in (28) and (30) are uniformly suffixational: in tuvisomavyo, the second vi‑ isn’t suffixed to the verb, but is instead (through aggregation) prefixed to the suffix ‑o; in a‑taka‑ye‑soma, ‑ye isn’t prefixed to the verb, but is instead (through aggregation) suffixed to the prefix taka‑. Swahili verb morphology is therefore canonical with respect to the affix directionality criterion.

We have already seen (Section 8.3) that the Swahili pattern of negative inflection apparently deviates from the rule opposition criterion, since the application of the first-person singular negative portmanteau rule ⟦si-⟧_III (as in si-tasoma ‘I won’t read’) is paradigmatically opposed to the application of the indicative negative rule ⟦ha-⟧_III together with a subject-agreement rule (e.g. with the second-person plural ⟦m-⟧_III rule in ha-m-tasoma ‘you (pl.) won’t read’) but that this deviation becomes merely apparent if we assume that what looks like the opposition of one rule to two rules is actually the opposition of one rule to another, composite rule (e.g to (⟦ha-⟧_III ◦ ⟦m-⟧_III)).

The patterns of relative inflection seen in this section also present a more extreme case in which one rule is opposed to combinations of as many as three rules; for instance, the past‑tense rule ⟦li‑⟧_II, the aggregated relativized‑argument rule (⟦vi‑₈⟧ Ⓐ ⟦‑o⟧_II), and the aggregated relative tense rule ((⟦vi‑₈⟧ Ⓐ ⟦‑o⟧_II) Ⓐ ⟦li‑⟧_II) all stand in paradigmatic opposition as members of Group II. Thus, the proposed analysis explains the relation of mutual exclusion that exists among simple or aggregated relativized‑argument rules, rules of tense and nonindicative negation, and aggregated relative tense/negation rules. A relative suffix never appears in postverbal position if there is overt morphology for tense or negation in preverbal position: this mutual exclusion is a consequence of the fact that the rules of relative suffixation in (28) and (30) belong to Group II, the same group as the rules of tense and nonindicative negation in (13b), (21b), and (28); all of these rules are in paradigmatic opposition to one another. To be sure, the rules of relative suffixation may combine with the rules of tense and nonindicative negation (as in Table 8.8), but their mode of combination is aggregation rather than composition; as a consequence of this aggregation, a relative affix appears preverbally in any verb form in which it coincides with the appearance of tense or negative morphology. Because the aggregations in Table 8.8 also belong to Group II, they are in paradigmatic opposition to the rules of relative suffixation in (28) and (30). This analysis therefore correctly accounts for the fact that the same verb form never has two relative suffixes, one in preverbal position and the other in postverbal position; this is a consequence of Pāṇini’s principle, which entails that the competition of an aggregated relative tense/negation rule with a (simple or aggregated) relativized-argument rule is always resolved in favor of the former rule. Finally, this analysis accounts for the fact that when they are in preverbal position, the relative affixes are always immediately preceded by a prefix expressing tense or negation: this is because relative affixes only end up in preverbal position as a consequence of the aggregation of a rule of relative suffixation with a rule of tense or negation, as in Table 8.8.

Interestingly, the analysis of Swahili verb inflection presented here derives additional support from the domain of phonology. As Barrett-Keach (Reference Barrett-Keach1986) has observed, polysyllabic verb forms in Swahili tend to consist of two phonological words, each with its own penultimate stress. The boundary between the two phonological words is after the tense prefix in (33a) and after the counterposed relative suffix in (33b).

In both of the instances in (33), the verb form’s first phonological word is a composite prefix defined by a composite rule of the type (R_III ◦ R_II): the composite rule (34a) defines the composite prefix ana- in (33a) and the composite rule (34b) defines the composite prefix anaye- in (33b). Thus, the interface of Swahili morphotactics with the phonology of the language may be assumed to conform to the general patterns in (35), however these are to be formulated.

Summarizing, the inflectional morphotactics of Swahili verbs presents a number of real and apparent deviations from canonical morphotactic criteria. Deviating from both the rule independence criterion and the unique sequence criterion, the argument‑coding rules in (14) depend for their application on their composition with the three carrier rules in Groups III, II, and I, and it is their composition with one or another of these rules that determines the morphosyntactic property that they express. The expression of negation deviates from the parallel sequence criterion. Both the formation of the relativized argument rules in (30) and the combination of these rules in Table 8.8 involve aggregation, and thus deviate from the stem operand criterion; at the same time, these aggregations reconcile both the formation and the combination of the relativized argument rules with the affix directionality criterion. The numerous apparent deviations from the rule opposition criterion are reconciled with it by the relations of composition and aggregation in the proposed analysis. At the most general level, all of the rules proposed in this analysis correlate with three groups, such that all of the relevant word forms are defined by FERs of either type (R_III ◦ R_II) or type ((R_III ◦ R_II) ◦ R_I).

The Swahili analysis developed here shows that the same basic rules may compose in different ways to realize different content. For instance, the composite rules in (36) involve the same basic rules but these are composed in different ways; rule (36a) is therefore the FER for the verb form wananipiga ‘they hit me’, while rule (36b) is the FER for the verb form ninawapiga ‘I hit them’. Although both of the composite rules in (36) involve the same basic rules, these are composed in different linear orders.

The next chapter focuses on a surprising characteristic of the Murrinhpatha system of verbal morphotactics: the fact that it allows the same basic rules to compose in the same linear order but with different hierarchical groupings to express different content.