1.

Introduction: Inflectional Distinctions

Most natural languages make inflectional distinctions: distinct forms of a lexeme reflecting the
different roles that it may play in syntax and in its contribution to the semantic composition of the
sentences in which it appears. Consider, for example, the forms of the French verbal
lexeme CONDUIRE ‘drive’ in (1). The forms conduisons and conduisez are used in the present
indicative, one with a first-person plural subject, the other with a second-person plural subject;
the forms conduisons and conduirons are both used with a first-person plural subject, one in the
present indicative, the other in the future indicative.

(1)

As these examples suggest, the distinctions among a lexeme’s inflected forms amount to
differences in morphological form that correlate with differences in morphosyntactic content
(differences in tense, person/number agreement, and so on). Thus, the inflectional distinctions
among the forms of a lexeme L may be represented in a table in which each cell (a)
corresponds to a unique combination of morphosyntactic properties with which L may be
associated in syntax and (b) houses the form of L that realizes that combination of properties. A
complete table of this sort is a representation of L’s inflectional paradigm; for
example, CONDUIRE has the inflectional paradigm in Table 1. Canonically, sameness or
difference of content among a lexeme’s inflected forms is invariably expressed as sameness or
difference in their morphology (cf. Brown, Chumakina, & Corbett, 2013; Corbett, 2005, 2009);
that is, a canonical paradigm has exactly one form per cell and a different form in each cell.
Nevertheless, deviations from this canonical pattern are extremely common; for instance, a
defective paradigm has one or more unrealized cells, an overabundant paradigm has alternative
forms in one or more cells, and a syncretic paradigm has the same form in more than one cell.

Table 1. The synthetic paradigm of French CONDUIRE ‘drive’

 The twin tasks of defining a language’s inflected forms and of identifying the grammatical
dimensions of the inflectional paradigms that house them raise of a number of fundamental
questions about the nature of morphology and its interfaces, which I address in turn in the
sections that follow:

 • What are the criteria by which inflectional morphology is distinguished from other sorts of
morphology, in particular derivational morphology (§2)?
 • What is the nature of the relation between an inflected word’s morphosyntactic properties and
the inflectional exponents of those properties (§3)?
 • Are inflectional exponents invariably synthetic markings, or can a lexeme’s inflected forms
include periphrastic expressions and host-clitic groups (§4)?
 • What is the most insightful way of modeling relations of inflectional exponence (§5)?
 • What is the canonical relation of content to form in a language’s inflectional paradigms (§6)?
 • What are the kinds of deviations from this canonical relation that a paradigm may exhibit (§7)?
 • How are such deviations significant for understanding the interfaces of inflection with other
grammatical components (§8)?
2. Distinguishing Inflection From Derivation
A long-standing concern in morphology is the need to differentiate between inflectional
distinctions among forms of the same lexeme and derivational distinctions among forms
realizing separate but related lexemes (Anderson, 1985; Booij, 2000; Stump, 1998, 2005). The
forms conduisons and conduisez are both forms of the verbal lexeme CONDUIRE, but the
form conducteur is customarily associated with a separate, nominal
lexeme CONDUCTEUR ‘driver’ deriving from CONDUIRE. Practical criteria such as those in (2) have
been invoked to differentiate between inflectional distinctions and derivational distinctions.

(2)
These criteria do, however, vary in their reliability. As it is stated, (2a) is a sufficient but not a
necessary criterion; derivation may fail to produce a change in syntactic category
(read ⇒ reread) or in lexical meaning (syntactic ⇒ syntactical). What’s more, (2a) is at odds with
the usual assumption that a verb’s participles and gerunds are among its inflected forms, even
though participles frequently have adjectival characteristics and gerunds, nominal
characteristics.

Criterion (2b) is substantially more reliable, though it, too, raises questions. For instance, the
distinction between different tenses is generally held to be an inflectional distinction, but in many
languages, different tenses aren’t obviously tied to different syntactic contexts. For instance, the
context John ___ cookies admits either bakes or baked. One might attempt to argue that the
sequence-of-tense phenomenon is a syntactic phenomenon that is sensitive to tense
differences, causing the tense of a higher clause to spread to a subordinate clause (as when
John is saying, “I bake cookies for a living” is reported as John said he baked cookies for a
living). But the sequence-of-tense phenomenon is less obviously a syntactic phenomenon than
a semantic or pragmatic one: John said he bakes cookies is also possible, and its lack of tense
sequencing affords it subtly different pragmatics.

Criterion (2c) is at most a statistical generalization. Defective paradigms are, precisely,

inflectional paradigms with gaps, and highly productive derivation may approach gaplessness.

Criterion (2d) often separates inflection from derivation along the same lines as other criteria,
but it doesn’t always. For example, there are some derivational operations that are extremely
regular in their semantics; an example is the derivation of ordinal numerals in -th. At the same
time, the phenomena of deponency (§7.4) and metaconjugation (§7.5) involve inflected forms
with unexpected semantics; somewhat less spectacularly, there are occasional instances in
which an inflected form is lexicalized with a special meaning, as in the case of brethren.

Criterion (2e) entails that in a word’s morphology, any inflection should constitute an outer layer.
On first consideration, this seems like an accurate generalization: In a complex word like norm-
al-iz-ation-s, the outermost suffix is inflectional, the remaining suffixes are derivational, and the
word as a whole is no longer open to additional morphological operations. The notion that
inflection follows derivation has therefore occasionally been accorded the status of a theoretical
principle, namely, the so-called Split Morphology Hypothesis (Perlmutter, 1988); ultimately,
though, this hypothesis proves to be empirically untenable (Bochner, 1984, Booij, 1994, 1996;
Stump, 1990A, 1990B). In fact, marks of inflection may appear internally to derivation in certain
circumstances.

First, category-preserving derivation gives rise to headed expressions (Stump, 1995, 2001, p.
96ff), and the head of such a derivative may be the locus of inflectional marking. For example,
Southern Barasano [Tucanoan, Colombia] has a category-preserving suffix -aka for deriving
diminutive nouns (Smith, 1973). This may apply both to ordinary count nouns (cotɨ ‘pot’
⇒ cotɨaka ‘little pot’) or to mass or collective nouns (hoa‘hairs’ ⇒ hoaka ‘little
hairs,’ oho ‘bananas’ ⇒ ohoaka ‘little bananas’); when one of these diminutives is inflected for
number (either through pluralization or singularization), the number inflection is internal to the
diminutive suffix, on the head noun (cotɨ-ri-aka ‘little pots,’ hoa-bą-ka ‘little hair,’ oho-ro-aka ‘little
banana’).

There are also instances in which category-changing derivational operations apply to inflected
forms; English has sporadic examples of this sort (worsen, betterment), but this can be a robust
phenomenon as well, as in Breton, where the denominal adjective suffix -ek applies to plurals
when it is semantically appropriate (e.g., drein-ek ‘thorny’ from drein, plural
of draen ‘thorn’; kerniell-ek ‘having horns’ from kerniel, plural of korn ‘horn’; preñved-ek ‘wormy’
from preñved, plural of preñv ‘worm’).
The less than full reliability of the criteria in (2) has led some morphologists to doubt the
theoretical validity of the distinction between inflection and derivation (e.g., Bochner, 1992, p.
12ff; Di Sciullo & Williams, 1987, p. 69ff). A more prevalent view is that inflection and derivation
are indeed distinct, but that this distinction only emerges clearly in the context of a carefully
articulated complex of assumptions about the architecture of grammar, the lexicon, and their
interface (Aronoff, 1994; Beard, 1995; Spencer, 2013; Stump, 2001, 2016).

3. Kinds of Inflectional Exponence

Exponence is the relation between an inflectional marking (an exponent) and the content that it
expresses. An exponent may be concatenative (e.g., the -ed in walk-ed) or nonconcatenative
(e.g., the ablaut in sing → sang). Logically, a variety of different kinds of exponence relations
are imaginable (Coates, 2000; Matthews, 1972). The simplest of these is biunique exponence,
in which (a) each exponent expresses a single morphosyntactic property and always expresses
that same property, and (b) each morphosyntactic property is expressed by a single exponent,
and always by that same exponent.

But relations between exponents and their content may also be one-to-many: In instances of
cumulative exponence, properties in different inflectional categories are systematically
expressed by a single exponent, for example, an exponent that expresses case and number
together. Exponence relations may also be many-to-one: In instances of extended (or multiple)
exponence, a word form has more than one exponent expressing the same content, for
example, a verb form in which past tense is expressed both by ablaut and by suffixation. In
instances of inflectional allomorphy, the same content is expressed by different exponents in
different word forms, for example, by ablaut in one word form but by suffixation in another; and
in instances of inflectional polyfunctionality, the same exponent expresses different but related
content in different word forms, for example, a first-person singular affix that codes a possessor
on nouns but expresses subject agreement on verbs.

An inflected word form’s inflectional exponence consists not only of its inflectional markings, but
of their ordering. Ordering relations among affixes are most directly observable, but even
nonconcatenative operations follow a particular order of application in their definition of an
inflected word form’s realization. The nature of such ordering relations (particularly as they apply
to affixes) is an object of vigorous debate. Can the relative ordering of affixes be deduced from
their semantic scope (Baker, 1985; Bybee, 1985; Rice, 2000)? Or are ordering relations
essentially arbitrary, describable only in templatic terms (Good, 2016; Stump, 1997)? Why are
the ordering relations among inflectional affixes rigid in most languages, but flexible in others
(Bickel et al., 2007; Luutonen, 1997; Noyer, 1994)? What is an appropriate model for the
representation of ordering relations (Crysmann & Bonami, 2016)? A comprehensive account of
these issues remains to be proposed.

4. Periphrasis and Clisis

Traditionally, inflectional exponents have been assumed to involve purely synthetic morphology;
thus, Table 1lists the synthetic forms of CONDUIRE. But the morphosyntax of French verbs also
involves periphrasis; for example, a verb’s ‘passé composé’ is formed by combining an auxiliary
(AVOIR ‘have’ or ÊTRE ‘be’) in the present indicative with the verb’s past participle: avons
conduit ‘(we) drove,’ sommes allés ‘(we) went.’ Is avons conduit one of the inflected forms
of CONDUIRE?
Recent research suggests that periphrases do in fact constitute inflected forms; for discussion,
see Ackerman and Stump (2004), Bonami (2015), Börjars et al. (1997), Chumakina and Corbett
(2013), Sadler and Spencer (2001), and Spencer (2003). One kind of evidence for this
conclusion is the complementary relationship that exists between synthetic morphology and
periphrasis in the morphosyntax of some category of lexemes. In some instances, the same
content that is expressed synthetically in the inflection of one subclass of lexemes is expressed
periphrastically in the inflection of another subclass. In Sanskrit, for example, it is usual for a
verb to inflect synthetically in the perfect tense; but certain verbs instead exhibit a periphrastic
inflection in which the verb, marked with a special suffix -ām, joins with perfect forms of the
verb AS ‘be’ (Stump, 2013). Table 2illustrates, with the synthetic perfect active inflection
of BUDH ‘awaken’ alongside the periphrastic perfect active inflection of BODHAYA ‘cause to
awaken.’ If BUDH and BODHAYA are both assumed to have perfect paradigms, then one must
assume that the paradigm of BODHAYA includes cells realized by periphrasis rather than by
synthetic morphology.

Table 2. Perfect active paradigms of two Sanskrit verbs

Although the complementarity of synthesis and periphrasis has a lexical dimension in this case,
there are other instances in which this complementarity has a purely morphosyntactic basis.
Latin verbs, for example, inflect both for perfectiveness (perfective or nonperfective) and for
voice (active or passive): Active forms are invariably synthetic, as are nonperfective forms, but
perfective passives are periphrastic; the paradigm of PARĀRE ‘prepare’ in Table 3 illustrates. As
this example suggests, all Latin verbs that have perfective passive cells in their paradigm realize
these cells periphrastically.

Table 3. Indicative forms of Latin PARĀRE ‘prepare’

A related issue concerns the status of clitics in inflectional morphology. In French, verbs serve
as hosts for pronominal and adverbial clitics: Jean vous y conduira ‘Jean will drive you there.’
Are host-clitic groups such as vous y conduira inflected forms of CONDUIRE? Recent proposals
for treating at least some clitics as inflectional morphology include Bonami and Boyé (2007),
Bonami and Samvelian (2008), and Miller and Sag (1997). As in the case of periphrasis, one
often finds a kind of complementarity in the use of synthetic morphology and that of clitics. In the
expression of prenominal possessives, personal pronouns inflect, usually suppletively
(I / my, you / your, we / our), but other nominals employ a phrase-final clitic (Dale’s, someone
else’s, a guy I know’s); here again, there is a lexical dimension to the alternation of synthetic
inflection with clisis.

But the complementarity between synthetic morphology and clisis may also be purely
morphosyntactic in its conditioning. Consider, for example, the paradigms of the Shughni
verbs WIRĪVDŌW ‘stand’ (intransitive) and WĪFTŌW ‘knit’ (transitive) in Table 4. In the present
tense, these verbs inflect alike, invariably exhibiting a person-number subject-agreement suffix.
In the past tense, the verbs inflect differently. The intransitive verb has two past-tense stems, a
default stem and a masculine singular stem; the transitive verb employs a single past-tense
stem that is insensitive to differences of gender or number. Person-number agreement is not
expressed suffixally, but by a second-position clitic situated on the first constituent of the clause,
whatever it might be; in these examples, it happens to be the subject pronoun. The intransitive
verb, unlike the transitive, lacks a second-position subject-agreement clitic in the third singular
of the past.

Table 4. Nonpast and past inflection of two Shughni verbs

It is clear that Shughni verbs agree with their subject in person and number; what is interesting
is that this agreement is expressed by synthetic morphology in the present but by a clitic in the
past. Thus, it is reasonable to think of the cells of past-tense paradigms as having realizations
that are most often discontinuous.

The notions of periphrastic inflection and discontinous realization present nontrivial problems for
understanding the interface of inflection with syntax; for discussion, see Bonami and Webelhuth
(2012), and Bonami (2015).

5. Modeling Inflectional Exponence

The simplest linguistic signs involve an arbitrary union of form and meaning. For example, the
union of the phonological form /dɔg/ with the meaning ‘canine’ is purely arbitrary, and must
therefore be stored in an English speaker’s memory. But words that are more complex are in
general not fully arbitrary; for example, the union of /dɔgz/ with ‘canines’ is motivated to the
extent that it parallels other form-meaning unions (e.g., that of /dɔg/ with ‘canine’ and that of
/bɝdz/ with ‘avians’). A perennial debate among morphologists concerns the most insightful way
to model such parallelisms in the form and content of inflected words. What, in particular, is the
most insightful way of modeling inflectional exponence—the relation between a word form’s
morphosyntactic properties and the morphological markings by which these properties are
expressed?

Structuralist grammatical theory developed the idea that a word w can be exhaustively
segmented into morphemes—nonoverlapping units such that (i) each unit has a form and a
content, (ii) no unit can be further subdivided into units having both form and content, and (iii)
the units’ form and content jointly constitute the form and content of w. The hypothesis that a
complex word can be modeled as the sum of its component morphemes has been remarkably
persistent.

Theoretical executions of this hypothesis have two important characteristics. First, they
are lexical in the sense that at the simplest level, associations between form and content are
stipulated in the lexicon as lists of morphemes; these basic associations then form the basis for
the associations of form and content arising through the combination of morphemes. Second,
morpheme-based theories of morphology are incremental: The form and content of a complex
word are built up cumulatively and in parallel, through the combination of its component
morphemes. In view of these characteristics, four kinds of phenomena constitute prima facie
counterevidence to morpheme-based theories of morphology; I shall label these Problems 1
through 4.

PROBLEM 1. The morpheme hypothesis’ focus on segmentability suggests that morphology is

fundamentally concatenative; thus, morpheme-based theories treat concatenative morphology
as primary and nonconcatenative morphology as a secondary effect triggered by a word form’s
concatenative inflection. For example, a morpheme-based analysis of English verb inflection
might assume that the -d suffix is the sole exponent of past tense in told, treating the vowel
substitution e → o not as an exponent per se, but as the effect of an operation of ablaut or stem
substitution triggered by the combination of tell with -d.

The assumption that nonconcatenative morphology is a secondary effect triggered by

morphemic exponents does present a problem for morpheme-based theories of inflection, since
some nonconcatenative operations are not overtly accompanied by a triggering morpheme; for
example, if the vowel substitution in told is triggered by -d, it is not apparent what triggers the
vowel substitution in broke. The gambit of postulating a zero past-tense morpheme to trigger the
vowel substitution in broke may solve the problem from a technical point of view, but it inevitably
raises issues of learnability. Why must a language learner postulate an inaudible exponent of
tense to trigger the vowel substitution in broke rather than simply treat the vowel substitution
itself as an audible exponent of tense?

More generally, the notion that morphology is fundamentally concatenative is at obvious odds
with the widespread incidence of nonconcatenative marks of inflectional content; for example,
concatenation of a noun’s stem with a plural morpheme is only one of the diverse ways in which
languages mark plural number (Table 5).

Table 5. Exponents of plural number in nine languages

PROBLEM 2. The morpheme hypothesis suggests that no morpheme in a word’s structure should
duplicate the content of one or more other morphemes in the same word, since that duplication
would be without motivation, failing to effect any increment in the word’s content; yet, extended
(or multiple) exponence—the incidence of two or more morphemes expressing the same part of
a word form’s content—is extremely common. The inflection of nouns in the Noon language
[Niger-Congo; Senegal] provides a clear example (Table 6). In Noon, nouns fall into six classes,
and a noun’s membership in a particular class determines the agreement morphology of its
modifiers. The nouns themselves inflect for number (singular or plural) and definiteness
(indefinite or definite), and definite forms additionally inflect for location (near the speaker, near
the hearer, or near neither interlocutor). Each location suffix is preceded by a class suffix
signaling the noun’s class membership; in addition, nouns belonging to classes 4–6 carry the
appropriate class prefix. Accordingly, forms such as p-ëlkít-p-ii ‘the thread near me’ exhibit
extended exponence, with the p- prefix and the -p suffix both expressing singular number and
class 5 membership. (See Harris, 2016, for extensive discussion of the phenomenon of
extended exponence, with abundant evidence from a range of languages).

Table 6. Indefinite and definite forms of nouns in the six Noon noun classes
PROBLEM 3. The morpheme hypothesis entails that no part of a word’s content should fail to be
associated with any of its individual morphemes. Yet, one commonly encounters instances of
underdetermination, in which some part of a word’s content cannot be directly attributed to any
one of its component morphemes. Consider, for example, the finite paradigm of the Bulgarian
verb KRAD ‘steal’ in Table 7. In this paradigm, krádox is unambiguously the first-person singular
aorist form, yet none of its component morphemes expresses either first person or singular
number; note, in particular, that the suffixes -o and -x both appear in all three persons of the
aorist plural.

Table 7. Finite forms of Bulgarian KRAD ‘steal’

PROBLEM 4. The morpheme hypothesis entails that inflected word forms have an internal
hierarchical structure. The incremental nature of morpheme-based approaches to inflection
relies on this structure to account for apparent scope relations among a word’s affixes; in
passive verb forms in Sanskrit, for example, the suffixal mark of passive voice is internal to (has
“narrower scope” than) the subject-agreement marker, whose agreement with a verb’s logical
object presupposes a passivized stem: dṛś-ya-nte [see-PRES.PASS-3PL] ‘they are seen.’ The
underlying assumption here is that inflected word forms are compositional in the same way that
phrases are—that just as the meaning of a phrase is a function of the meanings of its immediate
constituents, so the meaning of each complex constituent in an inflected word form’s
hierarchical structure is a function of the meanings of its immediate components (Baker, 1985).
Counterevidence to this assumption is, however, widely observable; in Latin, for example, the
verb form vide-nt-ur [see-3PL-PASS] ‘they are seen’ exhibits “scope relations” opposite to those
of Sanskrit dṛśyante, whose meaning it nevertheless parallels.

In view of these four apparent problems with the morpheme hypothesis, many morphologists
favor theories of inflection affording an account of inflectional content and form that is neither
lexical nor incremental. In lexical theories, the lexicon is regarded as an inventory of
morphemes, from which each morpheme is individually inserted into an inflected word’s
morphological structure. By contrast, inferential theories hold that the lexicon lists lexemes and
their stems, and that inflectional markings are not inserted directly from the lexicon, but signal
the application of rules that induce a lexeme L’s inflected word forms either from less complex
stems or from other word forms realizing L. Because the relations defined by these rules may or
may not be concatenative, they are compatible with the full variety of inflectional exponents,
avoiding Problem 1.

In incremental theories, a word’s content is built up cumulatively from the content of the
individual morphemes of which it is composed. By contrast, realizational theories hold that an
inflected word’s content logically precedes its form, serving in effect to drive the sequence of
rule applications that determine this form. In a realizational theory of inflection, it is perfectly
possible for the same part of a word’s morphosyntactic content to trigger the application of more
than one rule (in that way inducing extended exponence), and it is equally possible that part of a
word’s morphosyntactic content may fail to trigger the application of any rule (so that that bit of
content will be underdetermined by the morphology of the word’s form); in this way, realizational
theories avoid Problems 2 and 3. In general, realizational theories in no way entail that an
inflected word’s morphological structure should be isomorphic to its semantic representation;
 indeed, such theories are compatible with the view (Anderson, 1992) that an inflected word
form’s internal structure is nothing more than its phonological/prosodic structure (and thus avoid
Problem 4). Inferential-realizational theories therefore constitute a well-motivated alternative to
lexical-incremental theories of inflection based on the morpheme hypothesis.

Current models of inflectional morphology differ with respect to the lexical/inferential and
incremental/realizational dimensions (Stump, 2001). The family of lexeme-based models falling
under the “word and paradigm” label are in general both inferential and realizational. Among
these, some models induce a lexeme’s complex word forms “vertically,” from less complex
stems (Anderson, 1992; Matthews, 1972,); another induces a lexeme L’s complex word forms
“horizontally,” from other word forms realizing L (Blevins, 2006); and still other exploit both
vertical and horizontal realization (Stump, 2001, 2016; Zwicky, 1985,). Models such as those
described by Selkirk (1982) and Lieber (1992) are lexical and incremental. Steele (1995)
employs a model that is inferential but incremental. Distributed Morphology (Halle &
Marantz, 1993; Noyer, 1992) is lexical and in some ways realizational, though with incremental
undertones. It is realizational in the sense that the insertion of an inflectional exponent (or
‘vocabulary item’) into a word form’s constituent structure is not what contributes the content
that it expresses; that content is instead already present in abstract form in the structure into
which insertion takes place, having been integrated into that structure by syntactic principles. At
the same time, the assembly of a word’s morphosyntactic content by syntactic principles itself
has an incremental flavor to it. In word and paradigm theories, no such syntactic assembly is
necessary, since a word form’s morphosyntactic content is intrinsic to the definition of the
paradigm cell that it realizes; but typically of incremental theories, Distributed Morphology
attributes no theoretical necessity to inflection paradigms.

6. The Canonical Relation of Content to Form in

Inflectional Paradigms
Identifying a lexeme L’s inflectional paradigm is an intricate matter, presupposing the
identification of

 • the different morphosyntactic properties with which L may be syntactically associated and the
licit combinations into which such properties may enter; and
 • the complete inventory of forms that realize both L and one of the licit combinations of
properties with which L may be syntactically associated.
Corbett (2009) describes canonical inflectional paradigms as conforming to the systematic set of
criteria in Table 8. Column (a) specifies the canonical characteristics of an individual paradigm,
and Column (b), the canonical characteristics of a system of paradigms belonging to the same
syntactic category. By the first criterion, word forms within a paradigm are canonically alike in
their morphotactics (e.g., they all have the form Stem-suffix), as are word forms in
corresponding cells of distinct paradigms. By the second criterion, word forms within a paradigm
canonically share the same stem, while corresponding word forms in distinct paradigms
canonically have different stems. By the third criterion, word forms within the same paradigm
canonically differ in their inflectional marking (e.g., walk-s, walk-ing, walk-ed), while
corresponding word forms in distinct paradigms have the same inflectional marking (e.g., walk-
s, sleep-s, move-s). The net effect of these criteria is that canonically, word forms belonging to
the same paradigm are different, as are corresponding word forms belonging to distinct
paradigms.

Table 8. Canonical inflection paraphrased from Corbett (2009, p. 2)

If all inflectional paradigms conformed to these canonical properties, paradigms would be rather
uninteresting things—indeed, their theoretical significance could be reasonably called into
question. But inflectional paradigms are interesting and important precisely because they
deviate from these properties so often; indeed, one would be hard pressed to find a single
language whose paradigms didn’t deviate in various ways. These deviations, far from being
mere descriptive annoyances, provide important evidence for the place of inflectional
morphology in a language’s grammar.

7. Deviations From Canonical Paradigm Structure

Consider some examples.

7.1 Inflection Classes

Pattern (3) is a theorem of the canonical properties in Table 8. Many languages, however,
deviate markedly from this pattern. For example, among the Latin nouns in Table 9, no two
inflect identically in all 12 case-number combinations. These nouns are therefore traditionally
assigned to the four distinct declension classes indicated in the table. A declension class (like a
verb’s conjugation class) is an INFLECTION CLASS (Stump, 2015). If two lexemes inflect for
exactly the same morphosyntactic property sets, but employ different morphology to realize
these property sets, then the lexemes belong to distinct inflection classes. Thus, an inflection
class is customarily seen as a class of lexemes that are alike in their inflectional morphology but
that differ in their inflectional morphology from members of other such classes. Because
inflection-class distinctions are only important for a language’s morphology (being in effect
invisible for all other grammatical components), their incidence is one sort of motivation for the
view that, notwithstanding its interfaces with other components, a language’s morphology is
autonomous in its function.

(3)
Table 9. Examples of the five traditional Latin declensions

Although it is customary to think of inflection classes as classes of lexemes, they are more
plausibly seen as classes of stems. The inflection of a lexeme may involve more than one stem,
and these may or may not belong to the same inflection class. Consider the examples in
Table 10. In the paradigm of DĀTṚ ‘giver,’ the stems dātār-, dātar- and dātr- (with its
alternants dātur-, dātṛ- and dātṝ-) alternate according to the pattern of a single inflection class;
but the paradigm of KROṢṬṚ ‘jackal’ is heteroclite: Some of its stems inflect as members of the
class to which DĀTṚ’s stems belong, while others inflect as members of the distinct class whose
members include the stems of PAŚU ‘domestic animal.’ Thus, KROṢṬṚ is a single lexeme, but its
stems belong to more than one inflection class.

Table 10. The declension of three Sanskrit nouns

The heteroclisis of KROṢṬṚ is a lexically isolated, irregular phenomenon (Stump, 2006). But
there are also inflectional systems in which it is usual for a lexeme to have different stems in
different inflection classes (Stump, 2016). Sanskrit verbs, for example, generally inflect in four
different tense systems: present, aorist, perfect, and future. Each of the four systems has its
own peculiar set of conjugation-class distinctions: according to the traditional classification, the
present system has ten conjugations, the aorist-system seven, and the perfect and future
systems each two; see Table 11. Although there are occasional correlations, it is in general not
possible to deduce the class of a lexeme’s stem in one tense system from the class of its stem
in another tense system. Thus, it is essentially a matter of lexical stipulation that the
verb BHŪ ‘become’ has a present-system stem in the first conjugation (bháva-), an aorist-system
stem in the root aorist conjugation (ábhū-), a synthetic perfect-system stem (babhū́ v-), and a
future-system stem in the -iṣyá conjugation (bhaviṣyá-); that the verb STU ‘praise’ has a present-
system stem in the second conjugation (stā́ u-), an aorist-system stem in the iṣ-aorist
conjugation (ástāviṣ-), a synthetic perfect-system stem (tuṣṭā́ v-), and a future-system stem in
the -syá conjugation (stoṣyá-); and so on.

Table 11. Conjugation classes in the four Sanskrit tense systems

Inflection classes introduce an important dimension of complexity into a language’s inflectional
system. They engender what Ackerman and Malouf (2013) call the Paradigm Cell Filling
Problem—language learners’ task of deducing a full paradigm of inflected forms for a lexeme
only some of whose inflected forms they have actually encountered. In the simplest cases, a
lexeme’s full paradigm is directly deducible from the phonology of its stem; in Moru [Nilo-
Saharan; Sudan], for example, a verb’s conjugation-class membership is deducible from
whether its stem is monosyllabic or disyllabic and (if it is disyllabic) whether it is vowel-initial or
consonant-initial. But inflection-class membership may have grammatical correlates instead of
or in addition to phonological correlates; in Sanskrit, for example, the phonology of the noun
stem mātṛ- ‘mother’ limits its declension-class membership to a small number of choices, but the
choice among these depends on its feminine gender. Very often, however, a lexeme’s inflection-
class membership seems purely arbitrary. In English, for example, sit, slit, and knit belong to
three distinct conjugations: The three-way past-tense distinction sat, slit and knitted is without
rhyme or reason.

It has been widely hypothesized that there are limits on the difficulty of deducing a lexeme’s full
paradigm from a subset of its forms. According to the No-Blur Principle (Cameron-Faulkner &
Carstairs-McCarthy, 2000) and the Single Surface Base Hypothesis (Albright, 2002, 2008), a
paradigm’s inflection-class membership is generally deducible from a single diagnostic form;
there is, however, considerable prima facie counterevidence to this claim. Even so, a variety of
empirical measures consistently reveal important limits on the difficulty of filling in paradigm
cells (Ackerman et al., 2009; Milin, Kuperman, Kostić, & Harald Baayen, 2009; Moscoso del
Prado Martín, Kostić & Harald Baayen, 2004; Stump & Finkel, 2013).

7.2 Morphomic Properties

Inflection classes are an example of what Aronoff (1994) labels morphomes: categories that
have no grammatical or semantic significance beyond the confines of a language’s morphology.
Inflection classes are morphomes that have a lexical dimension: intuitively, they classify
lexemes according to the morphology by which their stems inflect for a shared system of
morphosyntactic property sets, contrary to the canonical property (3). But morphomes may have
a purely grammatical orientation, causing the realization of certain paradigm cells to conform to
a common pattern despite their morphosyntactic heterogeneousness. Morphomes of this latter
sort constitute a deviation from the pattern in (4), also a theorem of the canonical properties in
Table 8.

(4)

Consider an example of this sort of deviation from Bena-bena [Gorokan; Papua New Guinea],
whose verbs exhibit several suffix position classes. In the suffix position labeled ‘d’ in Table 12,
the default suffix is -be; but in the second-person singular and the first-person plural, -ne instead
appears. Thus, the content expressed by -ne is not morphosyntactically coherent, but is instead
morphosyntactically disjunctive. One might suppose that this is a matter of pure historical
accident, but other affixes exhibit this same disjunctive quality. For example, there is an
emphatic suffix that appears as -na by default (as in bila-na yabe ‘they will definitely go’) but as -
ta in the second-person singular or first-person plural (bila-ta yabe ‘you (sg.) will definitely go’);
similarly, there is an interrogative suffix appearing as -fi by default (as in bilu-fi-he ‘shall I go?’)
but at -pi in the second-person singular or first-person plural (bilu-pi-he ‘shall we (pl.) go?’). (See
Young, 1964, for additional discussion). It is clear that the -ne, -ta and -pi suffixes express a
morphomic property shared by second-person singular and first-person plural forms—a property
that has neither morphosyntactic coherence nor syntactic or semantic significance but which is
important for the language’s inflectional realization.

Table 12. Indicative paradigm of the Class A verb ho- ‘hit’ in Bena-bena (in compact and
exploded forms; a, b, c, and d label affix positions)
7.3 Defectiveness
In the canonical case, lexemes belonging to the same syntactic category have isomorphic
paradigms, each possessing the same number of cells, distinguished by the same system of
morphosyntactic property sets. There are, however, cases in which certain lexemes have a
smaller than usual number of cells (Baerman, Corbett, & Brown, 2010; Karlsson, 2000;
Sims, 2015). In some such instances, the deficit of cells has a reasonable semantic
explanation; scissors, for example, has no singular precisely because it refers to a joined pair of
blades. Yet, there are also lexemes whose cell deficit has no independent explanation—whose
paradigms are simply defective. A striking example is the paradigm of the French
verb TRAIRE ‘milk’ in Table 13.

Table 13. The paradigm of French TRAIRE ‘milk’

7.4 Deponency
Canonically, the same morphology recurs with the same function; in the inflection of English
verbs, for example, the suffix -s always expresses agreement with a third-person subject in the
present indicative. There are, however, cases in which the same morphology has one function
in one set of paradigms but a contrasting function in a different set of paradigms (Baerman et
al., 2007). In Old Norse, for example, the inflectional morphology exhibited in the past tense by
the strong verb FARA ‘go’ is exactly that exhibited in the present tense by the verb MUNU ‘will,’
whose own past tense matches that of the weak verb DŒMA ‘judge’; Table 14illustrates. The
verb MUNU ‘will’ is one of a small number of so-called preterite-present verbs, whose present-
tense inflections elsewhere serve as past-tense inflections. These preterite-present verbs
are DEPONENT (< Latin dēpōnere ‘lay aside’) in the sense that their present-tense morphology
has ‘laid aside’ its usual past-tense meaning.

Table 14. Indicative forms of three Old Norse verbs (Shaded forms and boxed forms are alike in
their inflectional morphology.)
7.5 Metaconjugation
A phenomenon related to deponency is what Stump (2016) calls METACONJUGATION. This is the
use of a particular inflection class to express different morphosyntactic properties in different
paradigms. In the inflection of Sanskrit verbs, for example, both the present system and the
aorist-system have preterite forms marked with the prefix a- (the so-called augment); these
preterite forms express the imperfect tense in the present system and the aorist tense itself in
the aorist-system. Strikingly, the thematic aorist-system conjugation employs the same
morphology in the aorist as the sixth present-system conjugation employs in the imperfect.
Thus, consider the verbs TUD ‘strike’ and TUṢ ‘be happy.’ The verb TUD has a present-system
stem in the sixth conjugation (augmented form atuda-) and an aorist-system stem in the s-aorist
conjugation (active form atauts-); the verb TUṢ has a present-system stem in the fourth
conjugation (augmented form atuṣya-) and an aorist-system stem in the thematic conjugation
(atuṣa-). As the forms in Table 15 show, the sixth present-system conjugation employs exactly
the same preterite morphology as the thematic aorist conjugation. That is, there is a single
conjugation that serves to express the tenses of the present system in the inflection of some
verbs, but instead serves to express the aorist tense in the inflection of other verbs; this
conjugation is labeled the sixth conjugation in the inflection of the first group of verbs and the
thematic aorist conjugation in that of the second group.

Table 15. The imperfect and aorist indicative active forms of two Sanskrit verbs
7.6 Syncretism
Canonically, every cell in an inflectional paradigm has a distinct realization, but very commonly,
different cells in a paradigm are realized by the same syncretic form (Baerman et al., 2005).
Syncretism can arise in more than one way; for example:

1. a) in some instances, the syncretized cells form a natural morphosyntactic class whose
syncretism reflects the lack of any morphology capable of distinguishing them;
2. b) in other cases, the syncretized cells stand in an asymmetrical relation such that one cell
takes on the realization proper to another cell; and
3. c) in still other cases, cells that do not constitute a natural morphosyntactic class share
their realization in a purely symmetrical way such that the shared realization is no more
proper to one of the cells than to another.
Syncretism of type (a) needn’t be stipulated, but can simply be attributed to a kind of poverty in
the resources available for realizing a word’s content. Syncretism of type (b) has sometimes
been attributed to rules of referral (Stump, 2001; Zwicky, 1985), which allow one cell to refer to
another cell for its realization. Syncretism of type (c) has a morphomic quality: The syncretic
cells are alike in possessing a purely morphological (= morphomic) property whose realization
they therefore all share.

Syncretism is a content-form mismatch of a rather distinctive kind: On one hand, it constitutes a

complication of the grammar, since it engenders ambiguity, at least potentially; on the other
hand, it constitutes a simplification of the morphology, requiring fewer forms to be learned for a
lexeme.

7.7 Conclusion
Together, these deviations from canonical paradigm structure—inflection classes, morphomic
properties, defectiveness, deponency, metaconjugation, and syncretism—suggest that a
language’s inflectional morphology constitutes a nontrivial interface of content with form.
Different approaches to inflection favor different ways of reading the implications of this interface
for understanding the architecture of grammar.

8. The Interfaces of Inflection

Among the most vigorous debates among contemporary morphologists is whether a language’s
inflectional system is simply a part of its syntax or is instead an autonomous grammatical
component whose interface with syntax is subject to tight restrictions. According to the
assumptions of Distributed Morphology, words have a hierarchical structure comparable to that
of phrases, and the shape of a word’s hierarchical structure is determined by independently
motivated principles for defining and reorganizing syntactic constituents. A word’s structure may
therefore comprise more than one terminal node in syntax; vocabulary insertion fills these
terminal nodes with roots and affixes, so that inflected words are literally formed in the syntax.

Lexeme-based approaches to morphology involve a markedly different conception of the

relation of inflectional morphology to syntax. In these approaches, a language’s morphology is
an autonomous system that defines a whole word form for insertion into a single terminal node
in syntax, subject only to the restriction that the morphosyntactic property set that conditions the
word form's inflectional realization be nondistinct from the property set that the syntax
associates with the node into which it is inserted. Word forms themselves have no internal
hierarchical structure other than their phonological/prosodic structure, and their syntax conforms
to the Lexical Integrity Principle (Bresnan & Mchombo, 1995).

A challenge for both of these conceptions of inflectional morphology is the wide range of ways in
which a language’s inflection may deviate from the canonical characteristics postulated in
Table 8. In Distributed Morphology, these deviations make it necessary to assume that a word
form’s syntactic structure undergoes various kinds of preprocessing in advance of vocabulary
insertion; these preprocessing operations such morphological merger, fusion, fission,
impoverishment, copying, and others (Embick & Noyer, 2001; Halle, 1997; Marantz, 1988;
Noyer, 1992).

In lexeme-based approaches to morphology, deviations from the canonical inflectional patterns

in Table 8 are seen as mismatches between the contentive aspects of a lexeme’s paradigm and
the inflectional realization of that content. Some such mismatches reflect a kind of poverty in a
language’s rules of inflectional realization—specifically, a failure to give morphological
expression to all of the content that is available for realization. Other mismatches reflect a kind
of parasitic dependency among paradigm cells, such that the realization of one cell simply
follows that of a contrasting cell. Still others involve morphology that realizes morphomic
properties, or the exceptional realization of some property by the morphology that ordinarily
serves to realize a contrasting property.

Stump (2016) advances the novel proposal that paradigms themselves are of two kinds: In a
lexeme L’s content paradigm, cells are distinguished by property sets that are accessible to L’s
syntax and semantics, while in a stem’s form paradigm, cells are distinguished by property sets
that are available for morphological realization. On this view, the cells in a lexeme L’s content
paradigm acquire their inflectional realization by virtue of their association with particular cells in
the form paradigm of L’s stem. Accordingly, a language’s inflectional morphology must specify
the mapping between the cells of a content paradigm and the associated cells in the
corresponding form paradigm; canonically, this mapping is trivial (with each form cell sharing the
morphosyntactic properties of the content cell that is associated with it), but each sort of
deviation from canonical inflection entails its own peculiar complication of this mapping