3.1 Introduction

Do paradigmatic gaps represent an anomaly – merely a superficial scar on the face of inflectional systems – or do they reflect something deeper and more systematic about grammatical structure? In this chapter an answer to this question is approached in terms of the question of why paradigmatic gaps occur in the first place and how they are related to principles of grammatical organization.

In sections 3.2–3.7, I give an overview of types of inflectional defectiveness, focusing on their causative factors. Previewing the discussion, the picture that emerges is one in which paradigmatic gaps have a wide range of causes and are rooted in interactions within and among many different areas of grammar. At the same time, a significant commonality that we find across various patterns of inflectional defectiveness is that they have a lexically specific element to their description, and in this sense are highly idiosyncratic. With regard to what paradigmatic gaps may be able to reveal about inflectional structure, a fundamental issue thus has to do with whether lexical idiosyncrasy (and concomitant item-by-item specification of defectiveness in the lexicon) should be equated with anomalies that lie outside of the “core” functioning of grammatical structure.

In sections 3.8–3.9, I use these data as a launching pad for a broad-scope discussion of this question. I argue that there is no fundamental conflict in the fact that paradigmatic gaps may be simultaneously lexically specified on an item-by-item basis and systematically organized according to regular grammatical principles. This idea draws on two different but compatible lines of thought. The first is the complex systems framework, which seeks to understand how relationships between elements in a system (including but not limited to language) give rise to the behavior and properties of the system as a whole (Evans and Levinson Reference Evans and Levinson2009a). A basic premise of complex systems research is that the properties of a complex system cannot be adequately understood by investigating the individual elements in isolation: the interaction between elements is crucial and the behavior of the whole is emergent from and not necessarily equal to the sum of the behavior of the parts. The second is a line of morphological research that has challenged the idea that there is a dichotomous split between irregularity and regularity and, correspondingly, between lexicon and grammar. Taken together, these lines of thought suggest the possibility of viewing the relationship between lexical organization and morphological structure as a dynamic interaction.

Before digging into the data, it is worth noting that the following discussion of the causes of defectiveness is meant to be more illustrative than comprehensive. Unfortunately, descriptive grammars do not always document inflectional defectiveness even where it does occur, and relatively few examples of defectiveness have come to light in the theoretical literature. In both kinds of sources, the focus has generally been on the discovery and elucidation of major, productive generalizations, rather than on the idiosyncratic marginalia. Moreover, paradigmatic gaps require the linguist to observe that some word-form does not occur, which is much more difficult than observing that a word-form does occur. For all of these reasons and probably others, we are only beginning to understand the ways that patterns of defectiveness vary from one language to another.Footnote ¹ Nonetheless, the attested instances are sufficient to show that the causes of inflectional defectiveness are wide and varied.

3.2 Defectiveness related to lack of semantic or pragmatic need

The most obvious (or, at least, most numerous) instances of missing inflected forms arise from a lack of syntactic, semantic, or pragmatic need. However, we might question whether these should be considered instances of inflectional defectiveness in the first place.

The definition of inflectional defectiveness that was offered in Chapter 2 intentionally excluded lexemes that have a restricted set of inflected forms because of a lack of syntactic need. But inflected forms commonly go missing because of a conflict between the lexeme's semantics and the meaning of a morphosyntactic feature value. To take just two examples, mass nouns like English information are incompatible with plural number (at least, without being divided up into quantifiable units, e.g., pieces of information), and inanimate nouns in Eastern Pomo systematically lack vocative case forms (McClendon Reference McClendon1977). The latter example can be understood as a function of the combination of inanimacy and the calling function of the vocative – people do not often talk directly to inanimate objects, certainly much less so than to animate beings. So while not in outright conflict, the combination is pragmatically incongruent. As Aristar (Reference Aristar1997: 316) notes, languages employ various “strategies for dealing with potentially type-incongruent cases and nominals. One is simply to disallow type-incongruent co-occurrences.”

If we take the view that syntactic grammaticality is independent of semantic interpretability, pragmatic appropriateness, or any issues related to context or patterns of usage,Footnote ² then we come to the conclusion that the plural of information in English and the vocative of inanimate nouns in Eastern Pomo meet the definition of paradigmatic gaps.Footnote ³ Even if contexts requiring the missing forms do not arise in practice, they could do so in principle.

However, semantically or pragmatically driven patterns of defectiveness are not particularly interesting examples of the phenomenon, at least not for the study of morphology. To the extent that forms are missing for semantic or pragmatic reasons, it is intuitively entirely expected that the relevant lexemes have a limited set of inflected forms. Whether a definition of inflectional defectiveness is formulated to include or exclude such examples, they are rather trivial examples of the phenomenon at best. Correspondingly, they receive very little attention in this book.

3.3 Defectiveness related to the morphology–phonology interface

Potentially more interesting from the perspective of morphological theory are instances of defectiveness that stem (in part) from a failure in the morphological form of the expected, but ultimately defective, word. Form-level failures have garnered the most attention in the theoretical literature on inflectional defectiveness, and several different ways in which the morphological form of a word can fail have been noted. In this section I present a few examples related to imperfect coordination between phonological and morphological systems. In section 3.4 I discuss form-level failures that derive from the inner workings of the morphological system by itself.

A fairly clear-cut example of phonologically motivated inflectional defectiveness comes from Swedish. Karlsson (Reference Karlsson2000: 648) notes that “in Swedish the ordinary genitive ending is -s. However, -s cannot be suffixed to stems ending in a sibilant. Nouns such as hus‘house’, svans ‘tail’, Max thus have no indefinite genitive singular forms like ^*hus-s, ^*Max-s (but they do have definite genitive singular such as bil-en-s, Max-en-s)” because the definite marker -en intervenes between the stem and the genitive marker, preventing the problematic phonological environment from occurring. The fact that the gaps appear only when the word-form would result in an illicit phonological sequence strongly suggests that the gaps are caused by a conflict in the interaction between morphological and phonological principles.

Similarly, Ito and Hankamer (Reference Ito and Hankamer1989) and Inkelas and Orgun (Reference Inkelas and Orgun1995) show that a handful of paradigmatic gaps in Turkish arise as a result of a minimum phonological length requirement. Some speakers of Standard Istanbul Turkish treat the words in (1) as ungrammatical, even though other forms of the same words are possible, as in (2). (Other speakers accept both as grammatical.)

1. (1)

   a.	*fa-m	‘musical note fa-1sg.poss’
   b.	*be-m	‘letter b-1sg.poss’
   c.	*ye-n	‘eat-pass’ (= ‘be eaten’)
   d.	*de-n	‘say-pass’ (= ‘be said’)

1. (2)

   a.	fa-mız	‘musical note fa-1pl.poss’
   b.	be-ler	‘letter b-pl’
   c.	ye-di	‘eat-pst’
   d.	de-mek	‘say-inf’

   (Inkelas and Orgun Reference Inkelas and Orgun1995: 770)

For those speakers for whom the words in (1) are ungrammatical, the problem lies in a general phonological requirement that words be at least two syllables long, and allomorphy does not apply to repair the phonological violation. Instead, the relevant paradigm cells are simply treated as defective.

Most examples in which phonotactic violations have been implicated as a cause of inflectional defectiveness are more complex, however, and involve a mix of phonological, lexical, and possibly morphological factors. An interesting if somewhat involved example comes from Hungarian, where a subset of verbs is defective in the definite and indefinite subjunctive, jussive, potential, and verbal adverb.Footnote ⁴ Lukács et al. (Reference Lukács, Rebrus and Törkenczy2010) argue that gaps in these paradigm cells arise from an interaction between phonotactic constraints, the inventory of stems available to a lexeme, and patterns of allomorphy among inflectional endings.Footnote ⁵

There are three crucial generalizations about Hungarian stem and suffix shape and stem distribution that determine the pattern of defectiveness. First, verbal stem allomorphs that end in a CC cluster sometimes have an “epenthetic” counterpart, i.e., an allomorph ending in CVC. The distribution of stem allomorphs is phonologically conditioned according to the following suffix: the CC-final stem is used when the following suffix is vowel-initial, and the CVC-final stem is used when the following suffix is consonant-initial. For example, mosolyog ‘smile’ appears as mosolyg-ok in the 1sg indefinite present tense indicative (‘I smile’), but as mosolyo̠g-hat-ok in the potential mood (‘I can smile’). However, some verbs have only a CC-final allomorph or only a VC-final allomorph. Lukács et al. (Reference Lukács, Rebrus and Törkenczy2010) refer to these as “stable” stems. Examples include kezd ‘start’, which has the indicative form kezd-ek ‘I begin’ and the potential form kezd-het-ek ‘I can begin’, and tanul‘study’, with the forms tanul-ok ‘I study’ and tanul-hat-ok ‘I can study’. Hungarian verbs can thus be classified according to whether they have stable stems or alternating stems.

Second, suffixes are also divided into stable types and alternating types, depending on whether the suffix has both C-initial and V-initial allomorphs, or only one or the other. Suffixes in the stable type include -ik (3sg indefinite indicative present), -ok/-ek/-ök (1sg indefinite indicative present), and -hat/-het (potential mood). Suffixes in the alternating type include -(a)nak/-(e)nek (3pl indefinite indicative present), -(o)tok/-(e)tek/-(ö)tök (2pl indefinite indicative present), and -(t)at/-(t)et (causative mood). The distribution of allomorphs for the alternating type of suffix is phonologically conditioned according to whether the end of the stem has a consonant cluster. In combination with stable stems we find, for example, kezd-enek ‘they begin’ and tanul-nak ‘they study’.

Third, in some CC-final stems – specifically, those in which the second consonant is either /l/ or /z/ – the cluster is not valid as a syllable coda. These all belong to the class of ik-verbs (e.g. omlik ‘collapse’).Footnote ⁶ In non-ik-verbs that have stems ending in a consonant cluster, that cluster is usually valid as a syllable coda and found in the language word-finally (e.g. hord ‘carry’). The ik-verbs that have stems ending in a coda-impermissible CC cluster may be either stable or alternating.

Collectively, these facts produce the conditions under which inflectional defectiveness occurs in Hungarian verbs. Specifically, paradigmatic gaps are found where a stable CC-final ik-verb pairs with a stable C-initial suffix. This produces a C[l,z]C consonant cluster at the morphological boundary which is disallowed phonotactically because the /l/ or /z/ cannot be syllabified either with the preceding syllable or with the following one. The result is a series of paradigmatic gaps. Examples of the expected but ultimately disallowed forms for the potential are shown in (3). According to Lukács et al. (Reference Lukács, Rebrus and Törkenczy2010), about sixty-five Hungarian verbs are defective in this way.Footnote ⁷

1. (3) Stable CC-final ik-stem + stable C-initial suffix = gap

   citation form	potential stem
   csukl-ik	*csukl-hat-	‘hiccup’
   meghasonl-ik	*meghasonl-hat-	‘become disillusioned, conflict with’
   özönl-ik	*özönl-het-	‘stream in large quantities’
   pirosl-ik	*pirosl-hat-	‘shine red’
   fogz-ik	*fogz-hat-	‘teethe’
   patakz-ik	*patakz-hat-	‘gush’

By way of contrast, some examples of the alternating type of ik-stems are given in (4).Footnote ⁸ These forms are unproblematic.

1. (4) Hungarian verbs with alternating ik-stems

   citation form	potential stem
   kotl-ik	koto̠l-hat-	‘brood’
   fesl-ik	fese̠l-het-	‘become unstitched’
   hajl-ik	hajo̠l-hat-	‘bend, incline, tend to’
   kibicsakl-ik	kibicsako̠l-hat-	‘become sprained’
   botl-ik	boto̠l-hat	‘stumble, trip’
   verz-ik	vere̠z-het	‘bleed’

Likewise, when a stable CC-final ik-verb is paired with an inflectional ending that begins with a vowel, this is also unproblematic. The second consonant in the stem-final cluster syllabifies as the onset to the following syllable (e.g., om.lik ‘he/she collapses’). It is thus clear that the phonotactic violation is a primary cause of this pattern of defectiveness because, as Lukács et al. note, the affected paradigm cells form a natural class phonologically, but not morphosyntactically.

Still, while the role of phonotactic violation in causing the Hungarian pattern of defectiveness cannot reasonably be denied, there are some complicating facts that do not fit neatly with a story about phonologically determined inflectional defectiveness. First, for some speakers “there are a few defective verbs that end in a cluster that does occur word-finally in a verb: e.g. /nz/ burjánz-ik ‘proliferate’ (compare the non-defective verb vonz ‘attract’)” (Lukács et al. Reference Lukács, Rebrus and Törkenczy2010: 87).Footnote ⁹ Phonological structure provides no explanation for such examples because there is no phonotactic violation at the boundary between the stem and the ending. Second, whether any given lexeme belongs to the alternating or stable type is a lexically specified property, apparently unpredictable from other elements of the grammar. The defectiveness that arises in non-alternating ik-stems thus has not only a phonological element to its causation but also a morphological/lexical one (belonging to the ik-class and the stable stem type). The need to acknowledge some degree of lexical specification is made particularly clear by the fact that a resolution strategy is readily available but not employed: alternation between CC- and CVC- stems (and between C-initial and V-initial suffixal allomorphs) already exists in the language. In principle, at least, the phonotactic violation that leads to defectiveness could be resolved by reanalyzing the relevant stems as alternating rather than stable. In short, the facts indicate that the Hungarian verbal gaps result from a combination of phonotactic, morphological, and lexical factors. A similar mixture of factors can be found in other cases of “phonotactic gaps” as well.

3.4 Defectiveness related to morphological structure

Problems with the shape of a word-form can arise entirely from within the morphological system as well, often as a result of competing inflectional rules.

For example, Munro (Reference Munro, Hardy and Scancarelli2005) connects paradigmatic gaps in Chickasaw to restrictions on affix combinability.Footnote ¹⁰ In Chickasaw, verbs inflect for the person and number of their arguments. There are three classes of argument markers; in both transitive and intransitive verbs these partly reflect the active/non-active status of the participant. In intransitive verbs, set I markers are associated with active participants (e.g. yaa-li [cry-1sg.I], ‘I cry’ [Munro Reference Munro, Hardy and Scancarelli2005: 124]) and set II with inactive participants (e.g. sa-kolofa [1sg.II-be.cut], ‘I am cut’ [Munro Reference Munro, Hardy and Scancarelli2005: 124]). When the verb is transitive, set I is used to mark the agent (subject), set II to mark a patient (object) or other inactive participant, and set III to mark the benefactive or other semantic experiencer (dative object).

However, which affix set a verb takes is also a lexical property, and as Baerman and Corbett (Reference Baerman and Corbett2010) observe, the affix sets behave like inflection classes in this respect; see (6).

1. (6)

   	set I	set II	set III
   1sg	hopoo-li	sa-chokma	an-takhoˈbi
   	‘I am jealous’	‘I am good’	‘I am lazy’
   1pl	(k)ii-hopoo	po-chokma	pon-takoˈbi
   	‘We are jealous’	‘We are good’	‘We are lazy’
   2	ish-hopoo	chi-chokma	chin-takhoˈbi
   	‘You are jealous’	‘You are good’	‘You are lazy’
   3	hopoo	chokma	in-takhoˈbi
   	‘He is jealous’	‘He is good’	‘He is lazy’

   (Baerman and Corbett Reference Lukács, Rebrus and Törkenczy2010: 7)

An important fact about Chickasaw in the present context is that no more than two overt argument markers can co-occur on the same verb, and one of them must be the subject.Footnote ¹¹ Moreover, having two argument markers from the same set is prohibited. This is usually unproblematic because the affix sets are tied to different semantic roles in transitive verbs. However, a small number of transitive verbs anomalously mark agents with set II markers. As Baerman and Corbett (Reference Baerman and Corbett2010: 7) note:

Similarly, set II and set III prefixes are compatible with set I markers, but sets II and III cannot co-occur. This combination results in defectiveness. “Is-sa-m-pilachi [2sg.I-1sg.II-dat-send] ‘You sent him to me’, with an unmarked third-person patient, is a fine sentence, but there is no simple way to express ‘He sent me to you’” (Munro Reference Munro, Hardy and Scancarelli2005: 126).Footnote ¹² Thus, the paradigmatic gaps in Chickasaw result from a failure in the form of the expected word, but for morphological reasons rather than phonological ones. Gaps arise where systematic restrictions on the combinability of affixes would be violated.

Hargus (Reference Hargus2007: 570) observes a similar kind of restriction in Witsuwit'en, whereby a certain derivational prefix combination “cannot be used in the imperfective.” It is an interesting example because the illicit forms (e.g. ^*Ɂəxwayeç ‘he is coming’) are well-formed when indicating customary actions, and are only defective when used as imperfectives. It is thus clear that the problem is not phonological. The data suggest that affix combinability offers partial explanation for the defectiveness of the imperfective, but there are other interacting factors as well.Footnote ¹³

Paradigmatic gaps have also been tied to competition among rules expressing the same set of morphosyntactic values. For example, Albright (Reference Albright, Garding and Tsujimura2003, Reference Albright2009) distinguishes two types of present tense indicative gaps in Spanish, what he calls “anti-stress gaps” and “anti-egotistic gaps.” Anti-stress gaps are illustrated in Table 3.1 on the left; anti-egotistic gaps are on the right. A few dozen verbs fall into one or the other type. Albright notes that the gaps follow the distribution of present tense morphophonological alternations. These are shown in Table 3.2, with diphthongization on the top left, raising on the bottom left, and velar insertion on the right. The gaps are distributed such that “anti-stress verbs are missing forms where diphthongization and raising occur, while anti-egotistic verbs are missing the form where velar insertion occurs” (Albright Reference Albright, Garding and Tsujimura2003: 4). Moreover, the defective lexemes belong to the inflection classes in which the alternations are possible but in which there is the greatest degree of variability in whether alternation applies within its conditioning environment. The defective lexemes also meet the structural conditions for (potentially) having these alternations. Thus, as Albright argues, the distributional facts strongly indicate a failure within the system that generates inflected word-forms.

He interprets the Spanish gaps as being caused fundamentally by uncertainty. The paradigmatic gaps occur disproportionately in low-frequency lexemes, which are not as likely to be lexically stored as high-frequency forms are, and which are thus more reliant on the inflectional rules of a language for their form. Moreover, the variable application of the alternation entails that no inflectional rule has high reliability in this context. Albright argues that taken together, these facts define a condition in which speakers are maximally uncertain about what the “correct” form of a word is, and the end result is a paradigmatic gap. Presumably, Spanish speakers avoid using forms that they are uncertain about, and at a societal level this manifests as a series of paradigmatic gaps.

But here, too, the full story seems to be complicated, and other researchers are more circumspect in their assessment of the same Spanish examples. Several criticisms of Albright's proposal have been put forward. First, there are verbs like blandir ‘brandish’ that show the same pattern of defectiveness as abolir, but which do not meet the structural conditions for diphthongization and raising. This word should thus not be subject to significant competition regarding stem shape. Albright's account incorrectly predicts such verbs to be well-formed. Second, Boyé and Cabredo Hofherr (Reference Boyé and Hofherr2010) point out that even where word-form uncertainty exists, this is not a problem in and of itself – speakers sometimes accept free variation. Third, Maiden and O'Neill (Reference Maiden and O'Neill2010: 108) note that the defective verbs are “relatively late, learnèd, loans from Latin or French” and that even in the paradigm cells where the verbs are not defective, they are “downright aberrant, contradicting an otherwise general principle of Spanish morphology.” Specifically, abolir and other defective verbs with back vowels in the stem fail to exhibit stem vowel alternation even where it is otherwise completely regular: abolir ‘abolish (inf)’, but anomalous abolió ‘sleep (3sg pst)’ (expected: abulió; cf. dormir ‘sleep (inf)’, durmió ‘slept (3sg pst)’). The implication is that the gaps and stem leveling are related strategies; defectiveness is a way to avoid a stylistic incongruity between the native Spanish patterns of alternation and the learned, borrowed nature of the defective lexemes. Albright's account offers no obvious way to connect this pattern of stem leveling to defectiveness in the same verbs.

Most importantly, however, both Maiden and O'Neill (Reference Maiden and O'Neill2010) and Boyé and Cabredo Hofherr (Reference Boyé and Hofherr2010) point out that even in the defective Spanish verbs like abolir and asir, more paradigm cells are defective than can be posited to have an active form-level conflict of the kind that Albright posits. In addition to the gaps in the present tense indicative that Albright is concerned with, the same verbs are also defective in the present subjunctive. Table 3.3 gives the full paradigm of abolir. While the present indicative gaps follow the distribution of an unpredictable alternation, the present subjunctive gaps do not. In the present subjunctive, the first and second person plural forms always have the same stem shape as the third person preterite, the imperfect subjunctive, and the gerund (Maiden and O'Neill Reference Maiden and O'Neill2010: 110). Since these paradigm cells are well-formed for abolir, we should expect the first and second person plural present subjunctive forms to be, without question, abolamos and aboláis, respectively. This undermines any attempt to explain the pattern of defectiveness purely in terms of conflicting inflectional rules.

Both Boyé and Cabredo Hofherr and Maiden and O'Neill argue that the generalization that unites the defective cells has instead to do with the morphomic distribution of stems. All of the defective cells would be expected to share a stem, based on independently motivated stem distribution patterns in Spanish.Footnote ¹⁴ A conflict that is (potentially) motivated in a subset of the present indicative cells is extended to all cells that share the same stem. This requires an additional generalization about the organization of stem space. “The type of defectiveness evidenced by the [Spanish] verbs…targets objects at the level of the stem space: certain stems are lexicalized as being absent” (Boyé and Cabredo Hofherr Reference Boyé and Hofherr2010: 45).

Thus, while morphophonological variation and the speaker uncertainty that it engenders almost certainly play (or historically played) a significant role in causing the Spanish gaps, it is equally clear that the two patterns of defectiveness also have lexical elements to their description, and other factors likely play a role as well. While Spanish is just a single example, a similarly complex mix of factors appears in Russian and Modern Greek, two languages that are discussed at length in the later chapters of this book.

3.5 Defectiveness related to morphosyntactic structure

Interestingly, inflectional defectiveness need not be related to a problem with the morphological form of a word at all. Sometimes the causes of paradigmatic gaps lie at the level of morphosyntactic features and their values. To see how morphosyntactic feature structure can lead to paradigmatic gaps, it is useful to take a diachronic perspective.

Mithun (Reference Mithun2010) argues that gaps in the paradigms of the Mohawk kinship terms -hsót and -ʼní are tied to how the language's gender system has developed historically. In Mohawk, all verbs bear prefixes indicating their core arguments; transitive verbs inflect for both agent and patient, realized jointly by a single prefixal morph. Most kinship terms are also morphologically verbs, so they use these same prefixes to denote who participates in the kinship relation. Interestingly, at least two kinship terms, -hsót ‘grandparent’ and -ʼní ‘parent’, have paradigmatic gaps when the “agent” is third person singular masculine and the “patient” is third person singular feminine-indefinite (3sg.m/3sg.f-indf); see (7). By contrast, all verbs and some other kinship terms (e.g., -htsi: ‘sibling’) having a well-formed word-form corresponding to this paradigm cell; see (8).

1. (7)

   a.	*shako-hsótha	‘he is grandparent to her’ = ‘her grandfather’	3sg.m/3sg.f-indf
   b.	*shako-ʼníha	‘he is father to her’ = ‘her father’	3sg.m/3sg.f-indf

1. (8)

   a.	shako-hsere’	‘he is following her’	3sg.m/3sg.f-indf
   b.	shako-htsi:'a	‘he is older sibling to her’ = ‘her older brother’	3sg.m/3sg.f-indf

Mithun argues that the historical failure of the feminine-indefinite to spread to all morphological verbs is implicated as a cause of the defectiveness of -hsót and -ʼní. Modern Mohawk expresses four genders, in the third person only. Proto-Iroquoian lacked a gender system, but had a third person indefinite category, used to indicate a generic person (‘one/people’). In the process of developing the system of gender oppositions found in the modern language, the third person indefinite morphed into the modern-day feminine-indefinite gender, used to refer to some female persons in addition to the generic function.

Significantly, however, distinct feminine-indefinite marking did not spread throughout the entire inflectional system. It seemingly never spread to some kinship terms that refer to the senior relation (e.g. -hsót ‘grandparent’, -ʼní ‘parent’). In the modern language, these lexemes when used to refer to senior female relatives (i.e., when the lexeme inflects for a feminine-indefinite third person singular agent) exhibit syncretism between feminine-zoic and feminine-indefinite gender.Footnote ¹⁵ However, kinship terms referring to senior male relatives (i.e., with masculine third person singular agents) are defective in all forms that require feminine-indefinite singular patient inflection, as shown in (7) above. As Mithun points out, there is very little synchronic rationale for the pattern of defectiveness in Mohawk. Moreover, from a historical perspective it seems most likely that the “missing” forms never existed in the first place. What changed is the expectation that there should be a form, as a result of the development of new morphosyntactic gender values.

Givón (Reference Givón1971: 413) famously said that “today's morphology is yesterday's syntax,” and syntax is certainly a major source of new morphosyntactic contrasts, alongside linguistic exaptation (Janda Reference Janda1996; Lass Reference Lass1990), and other kinds of reanalyses and extensions (Benveniste Reference Benveniste, Lehmann and Malkiel1968). But as Joseph (Reference Joseph and Siemund2011: 405) emphasizes, new morphosyntactic distinctions, whatever their source, are like other kinds of morphological change, in being “realized only in a piecemeal fashion.” The interesting thing about Mohawk is that the piecemeal development of the gender system results in distinct marking of the feminine-indefinite in some kinship terms, syncretism in others, and defectiveness in yet a third group.

Reflecting a somewhat different path of development but like Mohawk in being tied to the spread of morphosyntactic feature values to new lexemes, English beware has only one form, variously considered to be either a base form or the infinitive. Speakers categorically reject (9e) through (9h).

As Fodor (Reference Fodor1972: 531) points out, “[t]he real generalization about beware is that it can occur wherever uninflected be followed by an adjective can occur, e.g. in imperatives, infinitival complements, following modals.”Footnote ¹⁶

From a diachronic perspective, the defectiveness of beware has to do with the fact that the word derives jointly from two Old English two-word constructions: be + the verb warian ‘guard’, and be + the adjective waer ‘cautious’.Footnote ¹⁷ There is evidence of univerbation as early as 1300, from documents in which the construction is written as a single word (Oxford English Dictionary online: OED).Footnote ¹⁸ There are also some attestations of inflected forms (bewares, bewared, bewaring) in the seventeenth century, but these apparently never fully took hold and were ultimately rejected (OED). Why this happened is an interesting question – there is nothing phonologically or morphologically problematic about the forms in (9e)–(9h). But whatever the reason, the important thing in the present context is that, as in Mohawk, the defectiveness of beware primarily reflects a historical increase in what forms we expect to find, with no corresponding increase in the forms that we actually find, more than a dropping out of previously well-formed words.

Finally, the opposite of the Mohawk example can also be found, in which paradigmatic gaps arise during the demise of a morphosyntactic value. In Greek, the synthetic genitive form and a prepositional construction containing an accusative noun phrase compete.Footnote ¹⁹ The following examples are modified from Holton et al. (Reference Holton, Mackridge and Philippaki-Warburton1997).

1. (10)

   a.	Το	édose	tīs	Κaterínas
   	it	gave	the.gen.sg	Catherine.gen.sg
   	‘(S)he gave it to Catherine.’

   b.	To	édose	stī	Κaterína
   	it	gave	to.the.acc.sg	Catherine.acc.sg
   	‘(S)he gave it to Catherine.’

1. (11)

   a.	mia	megálī	merída	tou	laoú
   	a	large	portion	the.gen.sg	people.gen.sg
   	‘a large portion of the people’

   b.	mia	megálī	merída	apó	tōn	laó
   	a	large	portion	from	the.acc.sg	people.acc.sg
   	‘a large portion of the people’

Several factors are relevant to this variation. For instance, different semantic functions are amenable to the prepositional phrase to different degrees; Greek speakers typically prefer (10b) over (10a), but prefer (11a) over (11b). In some cases only the synthetic genitive form is normally possible, including in cases of possession (e.g., to spíti tīs_GEN Lukías_GEN ‘Lukia's house’; ^*to spíti apó tīs_ACC Lukía_ACClit: ‘the house from Lukia’), but prepositional constructions have been gaining ground over time. If current trends continue, the genitive will slowly die out as a case form in Modern Greek.

Based on my survey of two dictionaries (Babiniotis Reference Babiniotis1998; Lexikó tīs koinī́s neoellīnikī́s 1998), more than 2,000 nouns have paradigmatic gaps in the genitive plural (e.g., ^*kotṓn ‘chicken-gen.pl’ in (12a)).

1. (12)

   a.	*ta	podia	tōn	kotṓn
   	the	legs	the.gen.pl	chicken.gen.pl
   	‘the chickens’ legs’

   b.	ta	pódia	apó	tīs	kótes
   	the	legs	from	the.acc.pl	chicken.acc.pl
   	‘the chickens’ legs’

The gaps in the genitive plural seem to be connected to uncertainty about the form of the genitive plural; specifically, the gaps occur in those lexemes and in the paradigm cell for which the word-form is least predictable. (See Chapter 5 for detailed discussion.) However, they also seem to be connected to the more general pattern of genitive case loss. In Chapter 6 I argue that the development of paradigmatic gaps is facilitated by the existence of the largely synonymous and independently available prepositional phrase; it offers a way to avoid using genitive plurals about which speakers are uncertain. Example (12b) would normally be ungrammatical; this is one of the contexts in which the synthetic genitive is required. But because kóta ‘chicken’ is defective in the genitive plural, the prepositional phrase is used where it would otherwise not be possible. In this respect, the gaps may help to expand the semantic domain of the prepositional phrase, acting as a “wedge” from which case loss can further expand. Thus, while form-level competition among inflectional rules is certainly implicated as a cause of the genitive plural gaps, so is the broader morphosyntactic-level trend whereby the genitive is losing ground as a case value in Greek.

3.6 Patterns of defectiveness with no discernible causes

Finally, some patterns of defectiveness simply have no identifiable causes at all. Three short examples from a single language will suffice for illustration, although this is one area in which the data are robust.

A number of idiosyncratic patterns of inflectional defectiveness occur in Witsuwit'en, among different verbs and in different parts of the verbal paradigm (Hargus Reference Hargus2007). For instance, the verbs n-wət ‘be fast, walk fast’ and c’-ɬ-tsi ‘give birth’ both have only third person subject forms. Other person values are semantically viable and are apparently expected syntactically, but speakers circumlocute the missing forms using a syntactic construction.

1. (13)

   a.	nəwət	‘he's fast; he walks fast’
   	*nəswət	‘I walk fast’
   b.	c’əɬtsey	‘she gave birth’
   	c’əniɬtsey	‘it laid an egg’
   	*c’əsəcɬtsəy	‘I gave birth’

   (Hargus Reference Hargus2007: 569)

Also, the motion verb l-Geç ‘run’ has only imperative forms (second person imperfective). All of the extant forms of the verb are given in (14).

1. (14)

   selGec	‘hurry and run! (2sg)’
   səxwɬGec	‘hurry and run! (2du)’
   səxwɬɣez	‘hurry and run! (2pl)’

   (Hargus Reference Hargus2007: 571)

Finally, the verb ne#ɬ-ts'aç ‘return from expedition’ has only a single inflected form – third person imperfective negative.

1. (15)

   Q’ədəχ neweɬts'ac

   ‘He has still not returned.’

   (Hargus Reference Hargus2007: 570)

There is no need to belabor the point, but Hargus notes several more patterns of inflectional defectiveness among verbs in Witsuwit'en. In all, “[s]ome verbs cannot be inflected for first or second person, and some can only be inflected for second person. Some verbs have no perfective, some have no imperfective. Some have no positive forms. Some cannot be inflected for tense or subject” (573). Moreover, there is apparently no identifiable reason for these verbs to be defective, or for them to be defective in so many different ways. The causes of the gaps are obscure.

3.7 Interim summary

Even this rather incomplete survey of patterns of inflectional defectiveness is sufficient to show that the causes of defectiveness are wide and varied. A review of examples leaves the overall impression that every part of the morphological system, including its interactions with other components of the grammar, has the potential to give rise to defectiveness.

Defectiveness resulting from a lack of semantic or pragmatic need is rampant; for instance, it would not be surprising to find that every language that encodes nominal number has some singularia and/or pluralia tantum nouns. At the same time, such examples are rather marginal examples of the phenomenon. If inflectional defectiveness is the lack of an inflected form where we expect to find one, word-forms that are “missing” because of a lack of need for them violate expectations only in the most formal sense, and perhaps not even then.

Paradigmatic gaps can arise for reasons that have to do with morphophonological structure, or with morphosyntactic (and morphosemantic) structure of a language. Defectiveness arising from a form-level problem (e.g., violation of phonotactics or inflectional structure, or competition among inflectional rules) has garnered most of the recent theoretical attention, and for good reason, since this is where defectiveness is intuitively the most surprising. Phonotactic and morphotactic violations are so frequently resolved through allomorphy that it is surprising when defectiveness occurs instead. It is even more surprising when a pattern of allomorphy that could resolve the violation already exists in the language, yet fails to apply, as in the Hungarian data discussed above.

Finally, many examples of inflectional defectiveness have no discernible causes. In many cases, the original causes for the pattern are probably lost to history, and there is no synchronic causation to be found at all.

In their diversity, paradigmatic gaps offer an excellent way to peek into the functioning of inflectional systems. As Baerman and Corbett (Reference Baerman and Corbett2010: 11) aptly observe, “morphological systems, far from being streamlined media for implementing grammatical meaning, are ad hoc assemblages of imperfectly coordinated parts.” And by observing the myriad ways in which the coordination can be imperfect, we can see how that coordination is supposed to work.

3.8 Irreducible patterns of defectiveness and the inflectional system

The preceding discussion has emphasized a variety of causes of inflectional defectiveness, but a review of the literature also reveals a commonality within the diversity. It is striking how rarely paradigmatic gaps can be explained entirely by independent facts of the grammar. In many cases in which a synchronic, grammar-internal (or semantic) cause can be identified, there are additional, lexically specific elements to the story. For instance, in Hungarian (§3.3), defectiveness arises where a combination of stem and suffix would result in a phonotactically impossible cluster. A particularly interesting thing about the data is that similarly impossible clusters are frequently resolved in Hungarian via stem allomorphy or suffix allomorphy. Since stems and affixes are lexically specified as being either alternating or non-alternating (stable), we can view this as a competition between different principles of the language – maintaining the non-alternating status of a stem/suffix (a form of lexical conservatism [Steriade Reference Steriade1999]) versus the need for a word-form to express the intended meaning. It seems surprising that defectiveness would be preferred over introducing a morphophonological alternation that is novel for the particular lexeme but widespread in the language overall. The pattern of defectiveness in this language shows particularly clearly that paradigmatic gaps are never an inevitable outcome of morphological or phonological violation.

Even singularia and pluralia tantum nouns that are semantically rooted are still conventionalized facts of a language. For instance, information is a mass noun in English, but Croatian informacija ‘information’ is a count noun with both singular and plural forms (informacija (nom.sg), informacije (nom.pl)). In English, boot has both singular and plural forms, but in Nishnaabemwin, bootsan ‘boots’ (borrowed from English boots) is treated as inherently plural (Valentine Reference Valentine2001: 182). And so on. This is not to deny that there is some consistency in the kinds of concepts that show up as pluralia and singularia tantum nouns (see Karlsson (Reference Karlsson2000: 648–649) for a list), just as it is not to deny that phonological and morphological conflicts play a role in causing defectiveness. But it goes to show that despite the diversity of defective patterns, we almost always find what Stump (Reference Stump2010: 182) calls “irreducible patterns of defectiveness,” i.e., defectiveness that does not fall out automatically from independently motivated principles, and which therefore must be directly specified as part of a lexeme's morphological information, in whole or in part.

This should hardly surprise anyone who has given much thought to inflectional defectiveness. Still, the point bears discussion because, as was noted in section 1.2, lexical specification is often equated with ad hoc exceptionality to grammatical rules. Theories of language architecture have often relegated idiosyncrasies, especially among morphological patterns, to “the lexicon,” i.e., to a domain outside of the productive rules of grammar. To the extent that the description of most patterns of defectiveness has an element of lexical specification, this equation all but defines them as existing wholly separately from the system of inflectional rules. In an implicit but fundamental way, it serves to undercut the very possibility that (lexicalized) defectiveness could interact with inflectional rule systems in interesting ways.Footnote ²⁰ It is thus worth exploring this idea and its underpinnings in detail.

An example that we have already seen helps to illustrate the issues. Returning to the discussion of Spanish from section 3.4, remember that Albright (Reference Albright, Garding and Tsujimura2003) argues that the patterns of defectiveness found in Spanish verbs are a consequence of inherent, probabilistic competition between inflectional rules. He treats the Spanish paradigmatic gaps as fundamentally normal occurrences, in the sense that they are posited to be a product of the normal functioning of the Spanish inflectional system. At the same time, the gaps are an outcome of inflectional structure in his model, but they have no independent status. They are epiphenomena, not morphological objects.

Albright sees it as an advantage of his approach that he does not have to resort to lexical specification. He critiques earlier approaches to inflectional defectiveness, in particular, Halle's (Reference Halle1973) proposal that the Russian 1sg non-past gaps (see §1.1) be captured via a surface filter, formalized using a lexically specific feature: [–Lexical Insertion]. In Halle's model, the grammar generates the expected 1sg word-forms, but the [–Lexical Insertion] feature keeps those words from being inserted into syntactic structure. Albright (Reference Albright, Garding and Tsujimura2003: 12) argues that this approach predicts, contrary to fact, that paradigmatic gaps will be randomly distributed: “In principle, any form of any word could be marked as [–Lexical Insertion] or eliminated by a parochial constraint, but in fact, only certain forms, such as 1sg, are affected.” In essence, Albright argues here that lexical specification is merely stipulative.

The implicit logic that leads to this conclusion will probably be familiar to any linguist: whenever a word falls outside of the productive rules of grammar, it must be stored in the lexicon because it is no longer predictable. And conversely, words that are stored in the lexicon and recalled from there for use are not dependent on the productive rules of grammar for their formation. Lexical specification is thus equated with non-rule-generated form. Given that the primary goal of the field of linguistics is, arguably, the discovery and elucidation of productive, rule-governed principles of grammatical organization, it seems logical that lexical specification is something to be avoided wherever possible. The way in which Albright equates lexical specification of defectiveness with (a prediction of) random distribution through the lexicon makes sense only if the lexicon is the depository for everything which lies outside of the productive rules of grammar and is anomalous to the functioning of the grammatical system.

Albright has a valid point in that if paradigmatic gaps can be made to fall out epiphenomenally from inflectional rule competition, then such an approach is preferable to lexical specification. However, as discussed in section 3.4 above, Boyé and Cabredo Hofherr (Reference Boyé and Hofherr2010) and Maiden and O'Neill (Reference Maiden and O'Neill2010) both argue convincingly that this is not entirely possible even for Spanish verbal gaps, much less for all instances of inflectional defectiveness. The important questions in the present context thus become whether lexical specification really equates to “outside of the rules of grammar,” and whether lexically specified defectiveness is thereby implicitly relegated to the status of having, a priori, no possibility of significant interaction with the workings of the inflectional system.

At a basic level, Maiden and O'Neill (Reference Maiden and O'Neill2010) and Boyé and Cabredo Hofherr (Reference Boyé and Hofherr2010) conceptualize the relationship between inflectional defectiveness and the morphological system quite differently than Albright does. Their arguments that the distribution of gaps within a lexeme's paradigm conforms to the distribution of stems more generally leads naturally to thinking about the Spanish patterns of defectiveness as morphological generalizations in their own right that are integrated with inflectional structure by virtue of participating in the same morphomic patterns as well-formed inflected forms. This represents an approach to the relationship between the lexicon and productive rules of grammar that rejects the equation between being lexically specified and being outside of the rules of grammar.

These authors also treat paradigmatic gaps as fundamentally normal, but in a different sense than Albright does. For them, gaps are normal in the sense that lexically specified defectiveness reflects a generalization about the mapping between grammatical properties and inflectional form, not crucially different from other such generalizations. The Spanish gaps are morphological objects, and implicitly there is no more reason to expect gaps to be randomly distributed throughout the lexicon than there is to expect any inflectional formatives to be randomly distributed. Just as inflection classes have their own organizing principles and no linguist would probably ever suggest that the existence of inflection classes (i.e., the non-random distribution of inflectional formatives) must fall out epiphenomenally from inflectional rule competition, irreducible patterns of defectiveness can likewise have their own distributional principles.

The approach that Maiden & O'Neill and Boyé & Cabredo Hofherr take to the Spanish gaps parallels work on other kinds of morphological idiosyncrasy. To cite just a single line of thought that was also mentioned in Chapter 1, stem suppletion in Romance and other languages has been shown to follow the distribution of regular patterns of alternation, indicating shared, systematic principles of stem organization, despite suppletion being in other respects highly lexically idiosyncratic (Aski Reference Aski1995; Boyé and Cabredo Hofherr Reference Boyé and Hofherr2006; Hippisley et al. Reference Hippisley, Chumakina, Corbett and Brown2004). This work reflects a view of language that puts morphology at its center and that accords it an autonomous status. It rejects the idea that the lexicon is only a repository for idiosyncrasy and non-rule-governed phenomena. The implication is that there is no firm line between grammar and lexicon. The relationship between (lexicalized) inflectional defectiveness and inflectional structure becomes an issue for empirical exploration.

3.9 On causation and explanation

This raises the question of where we should locate causation and explanation, broadly speaking. The discussion in the early sections of this chapter focused primarily on structural, grammar-internal causation of the type typically preferred within the generative tradition. This was partly out of necessity – we simply do not know much about social factors as they relate to paradigmatic gaps, although there almost certainly is a social dimension at play. The discussion also gave somewhat short shrift to historical explanation because knowledge of the historical development of gaps is limited (but see Baerman Reference Baerman2008; Morin Reference Morin1987). Thus, while there is increasing understanding of the structural causes of paradigmatic gaps, external explanation (in the sense of Newmeyer Reference Newmeyer1998) is still largely a missing piece of the picture.

At the same time, the fact that so many examples of inflectional defectiveness have structural conditions that are necessary but not sufficient to their description, and the fact that defectiveness may never be the inexorable result of grammar-internal factors, suggests a need to move away from explanation purely in terms of structural causes. This involves a somewhat different mode of explanation from deterministic structural cause and structural effect. Specifically, it suggests explanation in terms of the dynamic properties of languages as complex systems.

This debate about the nature of explanation has been playing out in the field of linguistics in recent years. One instantiation of it consists of Evans and Levinson (Reference Evans and Levinson2009a), titled “The myth of language universals: language diversity and its importance for cognitive science,” no fewer than twenty-three short response pieces to that article that were published with it, and a response to the responses. Evans and Levinson (Reference Evans and Levinson2009b) respond to a variety of criticisms of their main article, including that their focus on surface diversity in language “leads away from the quest for general principles” (473). They argue forcefully that an evolutionary, dynamic adaptive systems framework of the sort that they promote is equally as interested in generality and explanation as a generativist framework. “But it differs in where we seek the general laws. For our generativist critics, generality is to be found at the level of structural representation; for us, at the level of process” (474).

In one respect this is an argument for external explanation over internal explanation, which has been a traditional divide between functionalist/typological/emergentist approaches and generative approaches, and they make the point that different evaluation metrics cause the camps to often talk past one another, rather than being in true dialogue.Footnote ²¹ Less frequently noted, however, is that Evans and Levinson are also making the point that there is a dimension to the question of explanation that divides the field but does not align entirely with internal versus external explanation, namely, representations versus processes. An approach that invests in evolutionary processes and the dynamic adaptive processes of complex systems extends beyond the question of internal versus external explanation by stepping back from assumptions about deterministic cause and effect relationships. While a complex system as a whole may still display deterministic behavior, complex systems approaches view this not as a direct relationship between individual elements in the system (e.g., representational structures), but more indirectly as a function of the dynamical processes of the system itself.

Evans and Levinson have a particularly low tolerance for abstractness of representation, and they are criticized for this in several of the response pieces. They are unconcerned with abstract grammar (and representational structure in general), presumably because in their view explanation and general principles lie in evolutionary processes. Ultimately, they are rightly criticized for all but completely dismissing the possibility of internal explanation. (For instance, a convincing case for the possibility of internal explanation is made by Perlmutter and Moore (Reference Perlmutter and Moore2002).) Nonetheless, the way that they shift the goal posts of explanation away from structural factors and towards the dynamics of complex systems is a move in the right direction.

They are certainly not the first to make such an argument; their ideas are rooted in a robust literature that extends well beyond linguistics to a number of fields that are concerned with problems related to how interactions between elements in a system give rise to the collective properties and behaviors of the system.Footnote ²² Nor is their paper necessarily the most representative example of the complex systems approach in linguistics. However, judging by the various response pieces, their articulation of the issues hit a nerve in linguistics. Part of this has to do with the somewhat different notion of explanation employed in the complex systems approach, as compared with more traditional modes of explanation in linguistics. Moreover, in such an approach, it is not only the explanandum that shifts, but also the explanans. Evans and Levinson (Reference Evans and Levinson2009a) argue that the unique elements of the world's languages should not be treated as marginal phenomena and deviations from some core, innate set of principles. Instead, both common and uncommon patterns are hypothesized to reflect the interplay of multiple causal factors, and the more unusual the pattern, the more it is potentially informative about the nature of this interaction. It is in this spirit that I take up the question of inflectional defectiveness.

A complex systems perspective suggests that while morphological structure may be “in the lexicon,” the lexicon should be investigated as a dynamic network. In this book I start from the position that inflectional defectiveness has the potential to offer a window into a dynamical interaction between the representational structure of the lexicon, learning and cognitive processing, and other systemic factors from which morphological structure emerges, and that the power of explanation is vested in this interaction.