2.1 Introduction
The definitional parameters of inflectional defectiveness are still largely unexplored territory; the first task is thus to establish foundational facts related to the phenomenon under investigation.Footnote 1 This is more than a mechanistic exercise in developing and applying criteria, however. This chapter offers a working definition, but as we will see, inflectional defectiveness needs precise definition exactly because the boundaries of the phenomenon are not clear. Defectiveness is a gradient phenomenon, and the nature of its gradience speaks empirically to the extent that it is (or more precisely, is not) anomalous to inflectional structure.
2.2 What it means for an inflected form to be “missing”
In section 1.1 I provisionally defined a paradigmatic gap as a “missing” word-form. This entails some expectation that a word-form should exist. It is therefore important that we define expectations in a meaningful way.Footnote 2 Here, it is useful to begin with a definition of the paradigm.
The inflectional paradigm is sometimes defined as the set of word-forms belonging to the same lexeme, but a better definition might be as the set of morphosyntactic and morphosemantic contrasts that a part of speech instantiates (e.g. Spencer Reference Spencer, Sadler and Spencer2004).Footnote 3 The latter definition is both more abstract, in the sense that it is one step removed from the actual word-forms that make up a lexeme's paradigm set, and it is arguably more precise, in that direct reference is made to the syntactic function of inflectional morphology. The syntax requires a given part of speech to have word-forms corresponding to certain sets of feature values, primarily for purposes of agreement and government.Footnote 4 A lexeme belonging to that part of speech has as many inflectional paradigm cells as there are well-formed, paradigmatically contrasting sets of values.
This definition of the inflectional paradigm gives us a principled way to identify paradigmatic gaps because it allows us to step back from morphophonological form and to thereby define what it means for a set of forms to be complete or incomplete. A set of word-forms belonging to some lexeme is complete if and only if, for every contrasting set of morphosyntactic and/or morphosemantic feature values that is syntactically well defined for that lexeme (i.e., for every inflectional paradigm cell), there is a word-form to express that set of values. A lexeme that does not have some word-form for every well-defined paradigm cell is incomplete. The overarching principle here is thus that the completeness of some set of word-forms must be defined in terms of the internal structure of the relevant language, and specifically, the shape of its lexemes’ inflectional paradigms. And paradigm shape is, in turn, defined primarily by the workings of the syntax.
Notice that being able to meaningfully define a set of word-forms as complete or incomplete is not the same thing as expecting it to be complete. The expectation of completeness derives from the nature of the morphology–syntax interface, as instantiated by the Principle of Morphology-free Syntax (Zwicky Reference Zwicky and Levine1992, Reference Zwicky, Brown and Miller1996).
The Principle of Morphology-free Syntax is a proposed universal principle of grammar. If we assume a generic feature-based syntax, it can be formulated as the idea that the syntax has access to morphosyntactic (and perhaps morphosemantic) feature values, but that it operates without regard for the morphophonological properties of words, including but not limited to inflection class membership. It is most naturally interpreted in the context of a grammar architecture in which (inflectional) morphology and syntax are generated by distinct grammatical components, with their interface mediated entirely through morphosyntactic features. As the same time, it is consistent with various theories of the morphology–syntax interface, including late-insertion models of transformational syntax, in which morphophonological information does not become available until a stage after the syntactic tree has been built and feature checking has operated, and declarative frameworks such as Head-driven Phrase Structure Grammar and Lexical Functional Grammar. It is thus pre-theoretic in nature.
Evidence for morphology-free syntax comes in part from the observation that if the syntactic system had access to morphophonological information, we should find syntactic rules that depend on morphological and morphophonological structure. Yet we never seem to find in any language syntactic rules that must be described as applying only to certain inflection classes, for instance, or as failing to apply to certain inflection classes.Footnote 5 Inflection class has no relevance outside of the morphological system. (For this reason, inflection class has been called a “purely morphological” feature [Kibort Reference Kibort, Kibort and Corbett2010], distinguishing it from morphosyntactic and morphosemantic features.) The Principle of Morphology-free Syntax encodes the idea that syntax is in this way systematically and universally insensitive to inflection class membership and other morphophonological properties of words. The syntax treats lexemes having the same morphosyntactic and morphosemantic properties but different morphophonological properties as equivalent. In essence, they are interchangeable cogs.
The expectation that all lexemes will have a full set of inflected forms can be seen as a corollary to the Principle of Morphology-free Syntax. For the sake of exposition, let us assume that when a lexeme is defective, as is the case for Russian verbs like pobeditˈ ‘to conquer’ (see §1.1), the word's lexical entry directly specifies that there is no word-form corresponding to some well-formed set of morphosyntactic values. We can thus think of information about a lexeme's defectiveness as being parallel to inflection class information – both involve lexeme-specific information about how morphophonological form expresses (or in the case of defectiveness, fails to express) morphosyntactic values. According to the idea of morphology-free syntax, we would thus expect the syntax to be blind to patterns of inflectional defectiveness. And indeed, this is what we seem to find.
More specifically, if information about inflectional defectiveness were available to the syntax, we would expect in some language to find syntactic rules that exist for the sole purpose of resolving the problem caused by failures of the morphological system. Instead, however, the circumlocutions that speakers employ always seem to take advantage of independently existing rules of grammar to express the same or a similar meaning, usually in an ad hoc way. (Remember that one way Russian speakers circumlocute the 1sg of pobeditˈ is by using the phrase oderžu pobedu ‘I will score a victory’ instead.) Thus, while speakers may be consciously or unconsciously aware that a given lexeme is defective, there is no evidence that syntactic structure is ever sensitive to the existence of a paradigmatic gap. Syntax seems to be blind to the possibility of inflectional defectiveness in accordance with the Principle of Morphology-free Syntax, and when the syntax requires some combination of lexeme and feature values that turns out to be morphologically defective, the morphological failure causes the syntactic structure to fail irreparably.
This is the sense in which we can derive an expectation that every combination of morphosyntactic feature values that is licit for a given part of speech will apply to all lexemes within that part of speech – an expectation of what Haspelmath and Sims (Reference Haspelmath and Sims2010: 93) term the unlimited applicability of inflectional values – and conversely, that every lexeme will have a complete set of licensed forms. Since the syntax is blind to morphological form, it requires lexemes to have complete sets of inflected forms in order to function smoothly. Moreover, while there is substantial fuzzy middle ground between inflection and derivation, any expectation that a derivational word family will be complete, or even any attempt to define what it would mean for a word family to be complete or incomplete, is on less solid footing because derivation is not tied in the same way to syntactic structure. The expectation of a complete set of forms is thus something special about inflection.
In fact, the completeness of the inflectional paradigm is such a deeply entrenched expectation that many theories treat it as a fundamental principle of inflectional structure, guaranteed via default rules. See, for instance, the treatment in Paradigm Function Morphology (Stump Reference Stump2001), where rules with more specific conditions apply to the exclusion of rules with more general conditions (a version of the Elsewhere Condition that Stump terms Pāṇini's Principle). The Identity Function Default – a rule that has no conditions, and which is therefore maximally general – guarantees that there will be a word-form for every licensed combination of feature values. While the details vary from one framework to another, default rules are widely used in inflectional theory.Footnote 6
So in summary, the syntactic function of inflection defines the shape of the inflectional paradigm in a way that is independent of the attested forms of a lexeme, thereby allowing a set of inflected forms to be identified as complete or incomplete. And the nature of the morphology–syntax interface creates an expectation that any lexeme will have an inflected form corresponding to each cell in the paradigm. It is exactly in this context that a violation of expectations – a “missing” word-form – is meaningful.
2.3 A working definition of inflectional defectiveness
We are now ready to formulate the definition in (1).
(1) A working definition of inflectional defectiveness
a. IF there exists a set of morphosyntactic and/or morphosemantic feature values F that is well-defined and morphologically encoded for at least one lexeme belonging to part of speech C;
b. AND IF there exists a well-formed syntactic structure S that requires F in combination with some lexeme L belonging to C;
c. BUT any form of LC that is inserted into S produces an ungrammatical construction;
d. THEN the paradigm cell defined by 〈LC,[F]〉 is defective.
In a language where both (1a) and (1b) are true, there is an expectation (in the sense identified above) that any given lexeme L belonging to part of speech C will have an inflected form to express feature value set F. I will refer to (1a) as the morphological expression requirement and (1b) as the syntactic need requirement. Point (1c) then identifies the conditions under which that expectation not being met constitutes inflectional defectiveness. I will refer to this as the ineffability requirement.
As a proof-of-concept illustration, we can see how the Russian example introduced in section 1.1 stacks up against this definition. As expected, the definition correctly identifies the 1sg non-past of verbs like pobeditʹ ‘conquer, defeat’ as defective. The feature-value set [tense: non-past, person: 1st, number: singular] is well-defined for verbs in Russian, and verbs generally have morphological expression of this set, as shown in Table 2.1.
Table 2.1 The Russian verb sprositˈ‘ask’
| sprositˈ ‘ask’ | singular | plural |
|---|---|---|
| 1st person | sprošu | sprosim |
| 2nd person | sprosišˈ | sprosite |
| 3rd person | sprosit | sprosjat |
Moreover, there are syntactic constructions that require the 1sg non-past form of pobeditʹ. Example (2) repeats (1) in section 1.1.
(2)
a. Kandidaty ot oppozicii pobedjat vo vsex pjati okrugax. candidates from opposition win.pfv.npst.3pl in all five regions ‘Opposition candidates will win in all five regions.’ b. *Ja pobežu vo vsex pjati okrugax. I win.pfv.npst.1sg in all five regions ‘I will win in all five regions.’
Finally, speakers deem any sentence that requires the 1sg non-past of pobeditʹ to be ungrammatical when filled with any form of the verb. The paradigm cell defined by 〈pobeditʹ, [tense: non-past, person: 1st, number: singular]〉 is thus defective.
This definition is, admittedly, imprecise in some aspects. For instance, words that syntactically behave like one part of speech but morphologically behave like another create a wrinkle. The discussion here pretends that “part of speech” is a theoretical primitive and that syntactic part of speech will match morphological part of speech. Mismatches occur, however; see Spencer (Reference Spencer, Orgun and Sells2005) for discussion. Where they arise, it is intuitively the syntactic part of speech that matters for present purposes. Still, since theories of morphosyntax have different approaches to formal specification of part of speech, the morphological expression requirement in (1a) is not be fully interpretable except in the context of a fully formalized theory.
Readers may also have noticed that the discussion so far has been vague about the level at which morphosyntactic and morphosemantic feature sets (F), and by extension the inflectional paradigm, are defined. Traditional grammars, not to mention many theoretical linguists, tend to assume that morphosyntactic feature values at the level of the morphology are in one-to-one correspondence with feature values at the level of syntax, and that there is thus no significant distinction to be made between these concepts. There is reason to reject this view, however. It is well documented that syntactic function and morphological form often exhibit different distributions. Spencer (Reference Spencer, Steinkrüger and Krifka2009: 200) states the issue clearly: “anyone who examines the morphosyntax of inflecting languages with reasonably rich case systems quickly concludes that there is a notion of ‘syntactic’ case to be distinguished from a purely formal notion of ‘morphological case’.” He argues for a two-level approach to case and other morphosyntactic features in which syntactic-level features (s-features) are visible to syntactic structure and are subject to semantic interpretation, whereas morphological-level features (m-features) are relevant only to determining morphological form. The mapping between the two may be isomorphic, but may also involve one-to-many or many-to-many correspondences and complex conditioning. See also Andrews (Reference Andrews and Bresnan1982), Maling (Reference Maling2001), Sadler and Spencer (Reference Sadler, Spencer, Booij and van Marle2000), Stump (Reference Stump2006), and Wechsler and Zlatić (Reference Wechsler and Zlatić2001) for similar argumentation.Footnote 7
Different approaches to analyzing morphosyntactic feature structure can affect whether a lexeme is determined to be defective; see the discussion of Finnish in section 2.5 below. But here it is sufficient to say that a conceptual distinction between s-features and m-features is useful, if for no other reason than that it allows the definition of defectiveness to be made more precise: the definition refers to s-features. While (1a) refers to morphological encoding, there is nothing to prevent the mapping between m-features and s-features to be non-isomorphic, so long as s-features (indirectly) receive morphological expression via m-features. Where the distinction between m-features and s-features is significant to the discussion in this book, I will assume such a two-level system of feature representation.
The overarching point here is that the interpretation of (1) may vary slightly from one theory to another. However, if we allow for intuitive notions of “part of speech,” “morphosyntactic feature,” and so on, it offers a working definition of inflectional defectiveness that operates on empirical grounds to the extent possible. The criteria in (1) define the empirical scope of inquiry of this book.
2.4 Drawing a line between defectiveness and other phenomena
It is important not only that the definition includes examples that intuitively fall under the umbrella of defectiveness, but that it excludes phenomena that are similar to, but not true examples of, defectiveness. This is accomplished primarily by including both a morphological expression requirement (1a) and a syntactic need requirement (1b). Defining the phenomenon jointly at both levels allows phenomena that are similar to true inflectional defectiveness at only one or the other level to be excluded. Examples that meet some subset of the criteria in (1) and/or which are (incorrectly) described in the literature as instances of defectiveness offer a useful test and clarification of the definition.
It is insufficient to define inflectional defectiveness purely in terms of syntactic need because we want to distinguish the lack of a syntactically required word-form due to morphological failure (= inflectional defectiveness) from the lack of a syntactically required word-form stemming from some other issue. For instance, syntactic constructions that require a target lexeme to simultaneously express two different values of the same feature arise in free relative constructions; see the Russian example in (3).
(3)
Ja ne mogla ponravitʹsja *komu/*kogo on nenavidit. I not could please who.dat.sg/acc.sg he hates. ‘I could not please the one whom he hates.’
The relativizer kto ‘who’ is required by a verb in the matrix clause to be in dative case, but is required by a verb in the subordinate clause to be in accusative case. Since it cannot express both simultaneously, the sentence is ungrammatical. A dummy to ‘that’ must be inserted in the matrix clause to express the dative argument, as in (4).
(4)
Ja ne mogla ponravitʹsja tomu, kogo on nenavidit. I not could please that.dat.sg who.acc.sg he hates. ‘I could not please the one whom he hates.’
Interestingly, however, constructions like (4) are grammatical if the relativizer happens to exhibit syncretism between the cases required by the two clauses, as in (5).Footnote 8 In Russian, čto ‘what’ exhibits nominative-accusative syncretism. It is thus able to simultaneously fulfill the requirement of the matrix verb for an accusative object, and of the subordinate clause verb for a nominative subject.Footnote 9
(5)
Ja kupila čto bylo v magazine. I bought what.acc.sg/nom.sg was in store. ‘I bought what was in the store.’
One way to interpret these data is that the problem with (3) has to do with the failure of kto to have a morphological form that fulfills the conflicting case requirements; (5) shows that a relativizer can fulfill arguments of both matrix and subordinate clause verbs. So does (3) show that kto has a gap in its paradigm?
In (3), there is some sense in which a morphological form is syntactically required but not provided, but we would not want to call this inflectional defectiveness. Rather, the morphological system is simply not structured in a way that can fulfill the syntactic need. With the exception of syncretic forms, which are at the very least a special case, the Russian morphological system does not have relativizers that simultaneously express two different combinations of case values. The ungrammaticality of (3) is thus not the result of morphological failure, exactly. In terms of the definition of inflectional defectiveness, we can see that the syntactic need requirement is met (1b), and the failure is of the right type (1c), but the morphological expression requirement in (1a) is not met. The definition is thus able to distinguish between true inflectional defectiveness and failures of the type that lead the utterance in (3) to be ungrammatical.
From the other direction, it is equally insufficient to define inflectional defectiveness purely in terms of morphological features and their expression because not all lexemes belonging to the same part of speech necessarily express the same set of m-feature values. Compare Russian sad ‘garden’ to zavod ‘factory’ in Table 2.2. Superficially, zavod looks defective in the second locative singular – it lacks a form that sad has.Footnote 10 Yet the definition in (1) does not classify zavod as having a paradigmatic gap in the second locative singular because zavod has a word-form available for every syntactic context in which it can appear. The syntactic contexts for the locative of zavod subsume the syntactic contexts of both the locative and the second locative of sad. Its syntactic behavior is thus quite different from that of, e.g., pobeditˈ (Table 1.1 in Chapter 1). A more traditional representation of the paradigm is as in Table 2.3. Since there is no syntactic failure in zavod, the ineffability requirement (1c) is not met, and this is an intuitively correct outcome. The second locative in sad is best treated as a subvalue of locative (Brown Reference Brown2007), with the difference between sad and zavod existing at the level of m-features, not at the level of s-features.
Table 2.2 Lack of expression of the second locative in Russian
| sad ‘garden’ | zavod ‘factory’ | |||
|---|---|---|---|---|
| singular | plural | singular | plural | |
| nominative | sad | sady | zavod | zavody |
| accusative | sad | sady | zavod | zavody |
| genitive | sada | sadov | zavoda | zavodov |
| dative | sadu | sadam | zavodu | zavodam |
| instrumental | sadom | sadami | zavodom | zavodami |
| locative | sade | sadax | zavode | zavodax |
| second locative | sadu | – | – | – |
Table 2.3 A traditional representation of Russian second locative as a subcase
| sad ‘garden’ | zavod ‘factory’ | |||
|---|---|---|---|---|
| singular | plural | singular | plural | |
| nominative | sad | sady | zavod | zavody |
| accusative | sad | sady | zavod | zavody |
| genitive | sada | sadov | zavoda | zavodov |
| dative | sadu | sadam | zavodu | zavodam |
| instrumental | sadom | sadami | zavodom | zavodami |
| locative | sade | sadax | zavode | zavodax |
| second locative | sadu | |||
One last issue has to do with what it means for some morphosyntactic value to be morphologically encoded. In Yimas, a language of the Ramu-Lower Sepik family, some prefixes (e.g., the negation marker) block others (e.g., nominative markers). The data get complicated, but (6) is a simple example in which the 2sg subject marker ma- appears in (6a), but not under negation in (6b) (taken from Wunderlich Reference Wunderlich2001: 349).
-
a. ma-ŋa-tpul
2sg.nom-1sg.acc-hit
‘You hit me.’
b. ta-ŋa-tpul
neg-1sg.acc-hit
‘You didn't hit me.’
Wunderlich describes taŋatpul as having a paradigmatic gap. Certainly, the example has some properties in common with gaps as defined here: the second person singular is well defined and (overtly) morphologically encoded for some words (e.g. maŋatpul), but there are other words in which the second person singular has no overt expression. However, the lack of an overt, segmentable morpheme expressing the second person singular subject is not the same thing as the second person singular subject having no morphological expression at all. Example (6b) is interpreted as having a second person singular subject. (It is also consistent with some other subject readings.) This is thus an example of zero expression of the nominative, which is not to be confused with lack of expression. Inasmuch as (6b) is a well-formed sentence and the ineffability requirement of the definition is thus not met, this does qualify as an example of inflectional defectiveness according to the criteria employed here.
Russian zero copula constructions are similar. A zero form is used in most predicative constructions that require a present tense copula, as in (7a).Footnote 11 By contrast, in the past and future tenses an overt copula is used, as in (7b,c).
(7)
a. On – vrač. he doctor.nom.sg ‘He is a doctor.’ b. On byl vračom. he be.pst.m.sg doctor.ins.sg ‘He was a doctor.’ c. On budet vračom. he be.fut.3sg doctor.ins.sg ‘He will be a doctor.’
Because there is a zero copula in the present tense, bytˈ is sometimes described as having a “defective paradigm” (e.g. Geist Reference Geist, Comorovski and von Heusinger2007: 81).Footnote 12 While it is true in a literal sense that there is no morphological form available for use in such constructions, this does not qualify bytˈ as defective according to the definition employed here, because the ineffability requirement is again not met. Sentences like (7a) are perfectly grammatical, despite there being no overt morphological form for the copula. In the end, there are various ways in which morphological expression can be lacking. However, these examples can and should be differentiated from inflectional defectiveness.
The proposed definition thus seems to serve its purpose. It includes phenomena that we intuitively identify as examples of inflectional defectiveness and excludes phenomena that are similar to defectiveness in some respects, but which ultimately represent distinct issues. Moreover, the discussion above highlights the extent to which inflectional defectiveness is a phenomenon that exists inherently at the intersection of morphology and syntax – an adequate definition must refer jointly to both levels of structure. Some phenomena that are not just mistaken for defectiveness but are truly on the boundary between defectiveness and other phenomena are discussed in the following sections.
2.5 The definition of defectiveness and the morphology–syntax interface
The definition of defectiveness proposed above assumes the prior analysis of morphosyntactic feature structure, yet it is not always entirely clear what the morphosyntactic structure of a given language is, because of different possible assumptions about how syntactic function and morphological form are related. So a question we should ask is whether this has practical consequences for whether a lexeme is identified as defective. Interestingly, it turns out that it can. Extended discussion of a single example from Finnish, based on Kiparsky (Reference Kiparsky2001), serves here to exemplify how different analyses of morphosyntactic feature structure can play out in the context of the definition of inflectional defectiveness.
As part of a broader argument about the relationship between abstract (syntactic) case and morphological case, Kiparsky (Reference Kiparsky2001) investigates a “small class of epithets” (321) in Finnish, including parka ‘poor’, that modify nouns and pronouns and must agree with their heads for case. He reports that both nominative and genitive syntactic contexts are unproblematic, as shown in (8a) and (8b), respectively.
(8)
a. Minä parka joudu-i-n siivoa-ma-an. 1sg.nom poor.nom get-pst-1sg clean-inf -ill ‘Poor me ended up cleaning.’ b. Minu-n para-n ol-i siivotta-va. 1sg-gen poor-gen be-pst.3sg clean-part ‘Poor me had to clean up.’
However, when the pronoun is in accusative case, some speakers report that there is no form of parka that makes the construction grammatical, as shown in (9a,b). The utterance in (9a) is marked with ‘%’ because it “is judged acceptable by some speakers and unacceptable by others” (Kiparsky Reference Kiparsky2001: 321), suggesting that there are at least two groups of Finnish speakers with somewhat different underlying generalizations. Here, we are only concerned with those speakers for whom the sentence is ungrammatical. (9b) is ungrammatical for everyone.
(9)
a. %He pan-i-vat minu-t para-n siivoa-ma-an. 3pl.nom put-pst-3pl 1sg-acc poor-?? clean-inf -ill ‘They made poor me clean up.’ b. *He pan-i-vat minu-t parka siivoa-ma-an. 3pl.nom put-pst-3pl 1sg-acc poor.?? clean-inf-ill ‘They made poor me clean up.’
Kiparsky explains the ungrammaticality of (9a) as resulting from the morphological feature structure of nouns and adjectives. In Finnish, nouns and adjectives inflect using the same set of case endings, with the exception of a small number of indeclinable adjectives (Sulkala and Karjalainen Reference Sulkala and Karjalainen1992). The traditional analysis of Finnish case structure treats the accusative as a nonautonomous case (Melˈčuk Reference Mel'čuk, Brecht and Levine1986), as shown in Table 2.4. (These are only partial paradigms. Finnish has many more cases than are shown here.) Personal pronouns decline similarly, but with an accusative form that is distinct from that of other cases. The case value of minut in (9) is thus unambiguously accusative (Table 2.5).
Table 2.4 A traditional representation of the partial case paradigm of Finnish nouns and adjectives
| äiti ‘mother’ | parka ‘poor’ | |||
|---|---|---|---|---|
| singular | plural | singular | plural | |
| nominative | äiti | äidit | parka | parat |
| accusative | äiti, äidin | äidit | parka, paran | parat |
| genitive | äidin | äitien | paran | parkojen |
Table 2.5 A traditional representation of the partial case paradigm of Finnish pronouns
| minä (1st person) | ||
|---|---|---|
| singular | plural | |
| nominative | mina | me |
| accusative | minut | meidät |
| genitive | minun | meidän |
Kiparsky argues that the crucial question has to do with what cases paran and parka express. He points out that if paran and parka are accusative forms that are syncretic with genitive and nominative, as the traditional analysis suggests, then at least (9a) should be grammatical. Moreover, the sentence is unquestionably grammatical when paran is removed, as in (10).Footnote 13
(10)
He pan-i-vat minu-t siivoa-ma-an. 3pl.nom put-pst-3pl 1sg-acc clean-inf -ill ‘They made me clean up.’
Kiparsky thus argues that the most logical conclusion is that paran and parka are actually morphologically genitive and nominative, respectively, and the examples in (9) are ungrammatical exactly because there is no morphologically accusative form of parka available, despite it being required by the syntax. In this view, the ungrammaticality stems from a failure of paran and parka to agree for case with minut. Example (11) repeats (9), but with glossing that reflects Kiparsky's analysis.
(11)
His alternative analysis of the case feature structure of Finnish nouns and adjectives is given in Table 2.6. The pronominal paradigm remains unchanged but in nouns and adjectives, abstract accusative maps either to morphological nominative or to morphological genitive. There is no accusative form. Notice that parka thus seems to have paradigmatic gaps in the accusative and Kiparsky identifies it as defective in this way.Footnote 14
Table 2.6 Partial case declension of Finnish nouns and adjectives, based on Kiparsky (Reference Kiparsky2001)
| äiti ‘mother’ | parka ‘poor’ | |||
|---|---|---|---|---|
| singular | plural | singular | plural | |
| nominative | äiti | äidit | parka | parat |
| accusative | – | – | – | – |
| genitive | äidin | äitien | paran | parkojen |
However, notice that if a noun replaces the pronoun in (9a), the construction is grammatical, as shown in (12). A traditional analysis labels äiti as syntactically accusative, based on the fact that in the same syntactic frame, pronominal forms are unambiguously accusative. And since parka is required to have the same case as the head noun, paran should also be construed as syntactically accusative. This reflects the analysis shown in Table 2.4 above.
(12)
He pan-i-vat äiti para-n siivoa-ma-an. 3pl.nom put-pst-3pl mother.acc poor-acc clean-inf-ill ‘They made poor mother clean up.’
Thus, in this analysis there is not only an accusative form of parka, but also a syntactic construction in which its use is grammatical.Footnote 15 Presented in this way, it seems that there is neither a failure in morphological expression nor an unmet syntactic need. Parka is thus not defective.
The crucial difference between this account and Kiparsky's is that the analytic principle employed in the latter does not result in äiti being classified as accusative in (12) (it is instead analyzed as nominative), so it and similar sentences are entirely beyond the scope of inquiry when it comes to the question of whether parka is defective in the accusative.
The traditional approach admittedly leaves unexplained why (9a) is ungrammatical for some speakers. However, notice that Kiparsky's analysis would entail labeling äiti as nominative and paran as genitive, so it seems to predict that (12) should be ungrammatical due to a failure of case agreement. Yet my Finnish informant produced (12) and found it to be entirely acceptable. (Crucially, she is in the group of speakers that judge (9a) to be ungrammatical.) So while it is hard to reject Kiparsky's core insight – that the failure of (9a) has something to do with the fact that pronouns have an explicitly accusative form but nouns and adjectives do not – neither the traditional approach nor Kiparsky's analysis is fully satisfying.Footnote 16
However, the point here is not to argue for one particular analysis of the case structure of Finnish, or even to argue that parka is or is not truly defective. Rather, the goal is to reinforce the idea, mentioned briefly already in section 2.3 above, that whether a lexeme is classified as defective can depend on fundamental analytic choices, particularly as related to the relationship between morphosyntactic features at the morphological and syntactic levels. Ideally, these principles of analysis are general and pre-theoretic, and not tied to the internal logic of any one particular theoretical framework. Situations in which the data look entirely different depending on analytic assumptions are probably not very common. Still, the fact that the definition of defectiveness presupposes morphosyntactic description means that it is not possible to identify paradigmatic gaps purely on empirical grounds.
2.6 Defectiveness and periphrasis
Even where we can establish that an inflected form is expected, it is not always easy to establish that it is missing. The boundary between defectiveness on the one hand, and periphrasis on the other, can be tricky.
2.6.1 Motivating a distinction
The first question to ask is whether there is an empirically grounded distinction between periphrasis and defectiveness in the best and most clear case. Consider the well-known example of Latin passive perfect periphrasis, shown in Table 2.7.
Table 2.7 Periphrasis in the Latin passive perfect of laudāre ‘praise’
| present tense | ||||
|---|---|---|---|---|
| active | passive | |||
| singular | plural | singular | plural | |
| 1st person | laudō | laudāmus | laudor | laudāmur |
| 2nd person | laudās | laudātis | laudāris | laudāminī |
| 3rd person | laudat | laudant | laudātur | laudantur |
| perfect tense | ||||
|---|---|---|---|---|
| active | passive | |||
| singular | plural | singular | plural | |
| 1st person | laudāvī | laudāvimus | laudātus/a/um sum | laudāti/ae/a sumus |
| 2nd person | laudāvistī | laudāvitis | laudātus/a/um es | laudāti/ae/a estis |
| 3rd person | laudāvit | laudāvērunt | laudātus/a/um est | laudāti/ae/a sunt |
In Latin, the passive mood has synthetic expression in the present tense, as does the active mood in both present and perfect tenses. However, the intersection of passive and perfect is expressed by a periphrastic construction – a participle (e.g. laudātus) that declines for gender and number plus an auxiliary formed from the copula (e.g. sum) that declines for person and number.
The Latin passive perfect exhibits both morphological and syntactic properties. The two elements of the construction are distinct syntactic units since they are separable and exhibit other word-like traits. At the same time, the passive perfect construction is morphological in the sense that it encodes independently motivated morphological features and stands in paradigmatic contrast with synthetic forms. Also, tense and mood cannot be doled out to either of the two parts of the construction; these values are necessarily interpreted as being carried by the construction in its entirety. The meaning of the Latin passive perfect is thus more than the sum of the meanings of its parts. Ackerman and Stump (Reference Ackerman, Stump, Sadler and Spencer2004) argue that semantic non-compositionality is a hallmark of morphology and one of the criteria that can be used to distinguish straightforwardly syntactic constructions from periphrasis.
So how do we draw the line between periphrasis and inflectional defectiveness? And should we? Playing devil's advocate for a moment, it is not difficult to imagine an alternative conceptualization in which the cells expressing passive perfect are instead treated as defective (shown schematically in Table 2.8). In this view, expressions like laudātus sum are simply syntactic circumlocutions used when a defective cell is encountered. Note that this would in many respects parallel the use of oderžu pobedu ‘I will score a victory’ as a circumlocution for the 1sg gap in the Russian verb pobeditʹ (§1.1). In both cases there is need for a form of the lexeme that realizes a set of well-defined morphosyntactic feature values, and a multi-word expression is employed when no one-word form is available. So should we conclude that the Latin verb laudāre is actually defective? Or alternatively, that the Russian verb pobeditʹ actually exhibits periphrasis and that defectiveness is an illusion?
Table 2.8 An alternate representation of the Latin paradigm of laudāre ‘praise’
| present tense | ||||
|---|---|---|---|---|
| active | passive | |||
| singular | plural | singular | plural | |
| 1st person | laudō | laudāmus | laudor | laudāmur |
| 2nd person | laudās | laudātis | laudāris | laudāminī |
| 3rd person | laudat | laudant | laudātur | laudantur |
| perfect tense | ||||
|---|---|---|---|---|
| active | passive | |||
| singular | plural | singular | plural | |
| 1st person | laudāvī | laudāvimus | – | – |
| 2nd person | laudāvistī | laudāvitis | – | – |
| 3rd person | laudāvit | laudāvērunt | – | – |
The Russian and Latin constructions differ in at least two important respects. First, the Latin periphrastic passive perfect has a fully conventionalized form – it always consists of a participle and auxiliary that decline in predictable ways. By contrast, the circumlocutions employed by Russian speakers are variable and are essentially ad hoc formations. Second, in Latin the periphrastic constructions that express the passive perfect are in complementary distribution with the synthetic constructions that express other feature value combinations. This kind of complementarity does not exist in Russian. A Russian speaker can choose to say either oderžišˈ pobedu ‘(you singular) will score a victory’ or pobedišˈ ‘(you singular) will be victorious’. Except in the first person singular, both types of expression are available, and they compete in some loose sense. The Latin construction laudātus sum is thus a grammatical realization of the lexeme laudāre in a way that the Russian phrase oderžu pobedu is not a grammatical realization of the lexeme pobeditʹ. And correspondingly, the lack of a synthetic passive perfect verb form in Latin is not a deviation from the general grammatical pattern of the language in the same way that the lack of a 1sg form of pobeditʹ is in Russian. Without pretending that either defectiveness or periphrasis is a homogeneous phenomenon, and allowing that the boundary between the two may be fuzzy (a point that I return to below), it is nonetheless clear that here as in many other cases there is a significant distinction to be made. But exactly how this difference is captured definitionally turns out to depend on major theoretical assumptions.
2.6.2 The distinction in theoretical terms
The dual nature of periphrasis has led to much debate about whether such constructions belong to the morphology or to the syntax. Brown et al. (Reference Brown, Chumakina, Corbett, Popova and Spencer2012) argue from a typological perspective that the question of whether periphrasis is morphological or syntactic constitutes a false dichotomy. They show that periphrases range in the extent to which they exhibit canonically morphological and/or canonically syntactic traits, but argue that this does not entail that periphrastic constructions fall along a one-dimensional scale with purely morphological properties at one end and purely syntactic properties at the other. Rather, periphrastic constructions may be both morphological and syntactic at the same time. However, until recently the literature has largely focused on the status of periphrasis as an either/or choice and the discussion here reflects that.
Some approaches have sought to locate periphrastic constructions solely within the syntactic domain, with complementarity between synthetic and periphrastic forms treated (in one way or another) as the result of blocking between the morphological and syntactic components (Ackema and Neeleman Reference Ackema, Neeleman, Legendre, Grimshaw and Vikner2001; Andrews Reference Andrews1990; Bresnan Reference Bresnan, Baltin and Collins2001; Kiparsky Reference Kiparsky, Booij and van Marle2005; Poser Reference Poser, Sag and Szabolcsi1992). This effectively treats all lexemes as lacking morphological forms exactly for those morphosyntactic values that are expressed by periphrasis instead. It seems initially to suggest that verbs like laudāre should therefore be classified as morphologically defective, along the lines of the representation in Table 2.8. However, a somewhat more careful look at the definition in (1) suggests a different interpretation. Remember that the Latin passive perfect always receives periphrastic expression. In such an approach, the construction is not treated as a morphological object, although the individual elements of it are. It thus seems that the morphological expression requirement (1a) would not be met. (Informally, we might say that the paradigm of laudāre is L-shaped rather than square.) In essence, if morphological expression is equated with being a single syntactic element, the morphology cannot fail in the case of the Latin passive perfect because it is not in the business of producing passive perfects in the first place.
Distributed Morphology offers a somewhat different possibility, in which constituent structure is manipulated post-syntactically but before vocabulary insertion so that a given set of morphosyntactic values surfaces sometimes synthetically and sometimes periphrastically (e.g., English comparative adjectives – happier vs. more wonderful) (Embick Reference Embick2000, Reference Embick2007). This approach is thus intermediary in the sense that it does not directly locate periphrasis either in the syntax or in the morphology (the latter of which itself is “distributed” throughout the grammar and not a distinct component). However, in terms of how the definition of defectiveness in (1) would apply, the issues are the same as with the blocking approaches.
Other proposals instead treat periphrastic constructions directly as exponents of inflectional paradigm cells (Ackerman and Stump Reference Ackerman, Stump, Sadler and Spencer2004; Börjars et al. Reference Baayen, Dijkstra and Schreuder1997; Sadler and Spencer Reference Sadler, Spencer, Booij and van Marle2000; Spencer Reference Spencer1999). They generally approach the complementarity of periphrastic and synthetic expression from an inferential-realizational perspective and thus emphasize that periphrasis belongs to the morphological domain, while not denying that there is a syntactic dimension to the phenomenon. For present purposes, the important feature of the morphological approach is that periphrastic and synthetic forms are both treated as morphologically generated realizations of sets of morphosyntactic values.
A particularly interesting and explicit version of the morphological approach is Bonami and Samvelian (Reference Bonami, Samvelian and Müller2009) (further elaborated in Bonami and Webelhuth Reference Bonami2013), which succeeds in capturing the intuition that periphrasis is both morphological and syntactic. They build an interface between the lexicalist syntactic framework Head-driven Phrase Structure Grammar (HPSG, Pollard and Sag Reference Pollard and Sag1994) and the inferential-realizational morphological framework Paradigm Function Morphology (PFM, Stump Reference Stump2001). There is not sufficient space here to go into details, but the importance of their model in the present context lies in the fact that periphrastic constructions are generated by realizational rules – the same mechanism used to generate synthetic forms. Equally importantly, the feature representation of a periphrastic construction is defined by the workings of the syntax; an HPSG sign of type word is well-formed only if the periphrastic form is licensed as the realization of the sign's morphosyntactic head features for the lexeme. If we were to extend this approach to the Latin passive perfect, it seems that laudāre would be classified as not defective according to the definition in (1) because the corresponding cells are formally filled with morphologically generated forms.
So in the end, all three approaches seem in the context of the definition to allow for the idea that periphrasis is conceptually distinct from defectiveness, at least in the case of Latin. But for different reasons.
2.6.3 The fuzzy boundary
While it is important to recognize the conceptual distinctness of inflectional defectiveness and periphrasis, there is also substantial middle ground where the distinction is far less clear-cut. Periphrasis is not a single, homogeneous type of construction but rather a motley collection of constructions that exhibit syntactic and morphological properties to different degrees (Brown et al. Reference Brown, Chumakina, Corbett, Popova and Spencer2012; Chumakina Reference Chumakina2011). For instance, periphrastic and synthetic expressions are not always in complementary distribution, nor do periphrastic constructions always have a fully conventionalized form. This can create a fuzzy boundary between periphrasis and defectiveness. For instance, if a periphrastic construction is used not only where an inflectional paradigm cell is motivated but no synthetic form exists but also as an alternative realization where a synthetic form does exist, then it is not entirely clear whether the cells for which there is no synthetic expression should be classified as defective. The partially overlapping distribution of periphrastic and synthetic constructions motivates an analysis of defectiveness in those cells where synthetic expression is unavailable, but an analysis as overabundance (Thornton Reference Thornton2011) is also possible.
The boundary between defectiveness and periphrasis often depends on the extent to which a morphological analysis of a periphrastic construction is motivated. Chumakina (Reference Chumakina2011: 263–267) discusses an example in Eastern Armenian that exhibits this type of issue. In Eastern Armenian, nouns referring to places can be expressed either with a synthetic locative form or with a postposition:
(13)
a. phoɣjch-um street-loc ‘in the street’ b. phoɣjch-i meǰ street-gen in ‘in the street’
For such nouns, the locative is preferred. However, for containers and other nouns that are not place names but can be used to indicate location, both types of construction are possible, but the postpositional phrase is preferred (e.g., tuph-i meǰ is preferred over tuph-um, both meaning ‘in the box’). In temporal expressions, the synthetic and periphrastic constructions apparently have different meanings, with the synthetic locative expressing a time interval, and the periphrastic indicating a point in time. Finally, “for the oǰ-declension, which comprises kinship terms and some other nouns denoting humans, it is claimed that there is no synthetic form of the locative at all” (Chumakina Reference Chumakina2011: 265). The (synthetic) forms of khyur ‘sister’ are given in Table 2.9.
Table 2.9 Marginal defectiveness in the Eastern Armenian kinship term khyur ‘sister’
| singular | plural | |
|---|---|---|
| nominative | khyur(ə) | khyurner(ə) |
| genitive | khroǰ | khyurneri |
| dative | khroǰ(ə) | khyurneri(n) |
| ablative | kroǰich | khyurnerich |
| instrumental | kroǰov | khyurnerov |
| locative | – | – |
Chumakina interprets the postpositional phrase as an instance of non-canonical nominal periphrasis whose distribution relative to synthetic locative forms is governed by semantic and lexical factors. While this analysis is well supported, it does not preclude an analysis that oǰ-declension is defective. The overlapping distribution of synthetic and postpositional phrase constructions in most nouns makes it less than clear that the postpositional phrase should be treated as realizing the locative paradigm cell. If it is not (and perhaps even if it is), nouns belonging to the oǰ-declension represent the failure of a (presumably syntactically licit) synthetic locative form that is otherwise motivated in the language. This is thus an instance of defectiveness, albeit more marginally so because of the ready availability of the postpositional phrase construction. In Chapter 6 I discuss the Greek nominal genitive plural, which exhibits some of the same issues, although more clearly landing on the side of there being genitive plural gaps in some noun classes.
The data thus show that the distinction between defective and well-formed morphological structures is not always clear-cut. Moreover, this is thus not simply a problem of analytic procedure but rather an issue of the empirical facts themselves. Well-formedness shades into defectiveness, undercutting the notions that defectiveness is something wholly different from well-formedness and that paradigmatic gaps are exceptional failures of inflectional structure.
2.7 Defectiveness and suppletion
The same issue arises when suppletive forms of the same lexeme are not in completely complementary distribution, for instance when there is overlapping suppletion. See Juge (Reference Juge1999) for examples of overlapping suppletion, and Juge (Reference Juge2013) for discussion of the historical relationship between suppletion and defectiveness. There is not always a clear choice between an analysis in which two semantically similar lexemes are posited, each with a defective paradigm, and an analysis as one lexeme with overabundance in some cells.
2.8 Gradient defectiveness
The definition offered in section 2.3 presupposes that the relevant paradigm cell is well defined, which is to say, it requires us to expect for some lexeme that we will find a word-form corresponding to some set of morphosyntactic and/or morphosemantic values. However, expectations are gradient, having to do with how well integrated morphosyntactic and morphosemantic values are into the inflectional system of a language as a whole, and about variability across features and languages in this respect. Correspondingly, the line between defectiveness and well-formedness is gradient rather than binary.
2.8.1 Gradient expectations
Corbett (Reference Corbett, Corbett and Noonan2008, Reference Corbett2011, Reference Corbett, Brown, Corbett and Chumakina2012) outlines a set of canonical traits of morphosyntactic values. Exhaustiveness, having a distinct form, and unlimited combinability are three of these, alongside others. He argues that the full set of traits may or may not be observed in any natural language, but by mapping real-world examples onto the hypothetical space defined by the canonical traits, it is possible to quantify how close some particular instance comes to being an example of a canonical feature value. Looking at morphosyntactic case in particular, he argues that the advantage of this approach is that it “allows us to handle gradient phenomena in a principled way … [W]e can locate instances of case values as more or less canonical, rather than having an ‘all or nothing’ requirement that a particular instance is a case value or it is not” (Corbett Reference Corbett, Corbett and Noonan2008: 4).
Only a subset of the issues discussed by Corbett are of direct relevance here, but in the spirit of the Canonical Typology approach, we can identify two issues of interest: unlimited applicability (a similar but somewhat narrower notion than exhaustiveness) and unlimited combinability.
(14) A morphosyntactic or morphosemantic value exhibits unlimited applicability if it receives formal expression in all parts of speech that encode that feature, and within these parts of speech, in all lexemes.
(15) A morphosyntactic or morphosemantic value exhibits unlimited combinability if it receives formal expression in combination with all values of other morphosyntactic and morphosemantic features that are applicable to the same lexeme.
A morphosyntactic or morphosemantic value is more canonical if it exhibits greater applicability and greater combinability. Taken together, applicability and combinability are a measure of how integrated a morphosyntactic or morphosemantic value is into a language's system.
This dimension is important because morphosyntactic and morphosemantic values differ widely in how well integrated they are. Without trying to do justice to the wide range of possibilities, it is sufficient for the present purposes to think in terms of a loose scale from those morphosyntactic values that are most integrated to those that are least integrated, as measured by how widely they apply to the lexical stock and how fully they combine with the values of other features. At one end of the hypothetical scale, farthest from canonical behavior, are morphosyntactic and morphosemantic values that are restricted to only one part of speech and to only a subset of lexemes within it (very limited applicability). The example of the Russian second locative that was discussed in section 2.4 reflects this situation.
One step closer to being integrated, Corbett (Reference Corbett2000: 92) notes that in Yimas, personal pronouns mark four number values: singular, dual, paucal, and plural. Nouns also mark number, but only singular, dual, and plural.Footnote 17 Paucal number is never morphologically expressed on nouns. So unlike with Russian locative, paucal applies to all lexemes in one part of speech (if we consider personal pronouns to be distinct in this respect from nouns). But paucal is still not very integrated into the inflectional system of Yimas, given that the value does not apply to all parts of speech that mark number.
Albanian case and definiteness marking is yet one step closer to the integrated end of the scale with regard to feature value applicability. In this language, nouns inflect for case, number, and definiteness, and have gender as an inherent property. Particles of concord (e, i, tё, sё), which link the noun and adjective and which also serve other functions in Albanian grammar, agree with nouns for all four features.Footnote 18 However, when adjectives occur in their typical postnominal position, they inflect only for gender and number.Footnote 19 Example (16) shows the invariance of the adjective across both definiteness and case, and Table 2.10 gives the full inflectional paradigm of the adjective mirё ‘good’ in postnominal position (minimal though it is).
(16)
a. vajzё e mirё girl(f). nom.sg.indf f.nom.sg.indf good.f.sg ‘a good girl’ b. vajza e mirё girl(f).nom.sg.def f.nom.sg.def good.f.sg ‘the good girl’ c. vajzё tё mirё girl(f).dat.sg.indf f.dat.sg.indf good.f.sg ‘(to) a good girl’ d. vajzёs sё mirё girl(f).dat.sg.def f.dat.sg.def good.f.sg ‘(to) the good girl’ (modified from Newmark et al. Reference Newmark, Hubbard and Prifti1982: 183)
Albanian case and definiteness values are more integrated into the inflectional system than are Yimas number values because case and definiteness values are marked consistently within the classes that inflect for those features. At the same time, case and definiteness values are not fully integrated. This is made particularly clear by the fact that we might reasonably expect adjectives to inflect for case and definiteness, given that they agree with nouns for gender and number, and other agreement targets (i.e., particles of concord) mark both case and definiteness. But adjectives systematically fail to express these features.
Table 2.10 Albanian adjectival declension
| mirё ‘good’ | singular | plural |
|---|---|---|
| masculine | mirё | mirё |
| feminine | mirё | mira |
At the very canonical end of the scale, there is probably no language that expresses some morphosyntactic feature for all parts of speech, or even all content parts of speech. However, there are languages for which the distinctness of the categories noun, verb, and especially adjective is questionable because of strongly similar morphological and syntactic behavior. Some languages of the Wakashan, Salishan, and Iroquoian families, for instance, seem to distinguish nouns and verbs only in a limited way (Haspelmath Reference Haspelmath, Baltes and Smelser2001: 16543; Jelinek and Demers Reference Jelinek and Demers1994). A more common scenario is for the distinctness of adjectives to be in question (Dixon Reference Dixon1977); for instance, Korean words that express properties behave like verbs (Haspelmath Reference Haspelmath, Baltes and Smelser2001: 16542). Languages where the distinctness of content parts of speech is in question do not entirely mark every feature on every part of speech, but they come closer to this than the typical Indo-European language does. The degree of integration of some feature value into a language's inflectional system is thus clearly gradient. A similar gradience can be identified with regard to feature combinability.
We transform this into a statement of expectations by equating degree of integration of some feature value with strength of expectation of a word-form expressing that value. This is not much of a leap; the notion of expected behavior is already implicit in Zaliznjak's (Reference Zaliznjak1967, Reference Zaliznjak and Zaliznjak1973) discovery procedures for identifying case and other morphosyntactic feature values (an English summary is available in Corbett Reference Corbett2011). Imagine a simple scenario in which noun lexeme A and noun lexeme B can both occur in syntactic contexts 1 and 2 (e.g., as direct object of a transitive verb, and after a preposition). In both of these contexts, lexeme A has morphological form a, but lexeme B has a different form in each. According to the basic procedure proposed by Zaliznjak, we identify lexeme A as realizing two different case values despite it having only one form (i.e., we analyze this as case syncretism) exactly because in lexeme B, two different forms are used in the two different contexts. Cross-lexeme comparison is thus crucial to the procedure; expectations and conclusions with regard to A are derived from observation of B. The more restricted some morphosyntactic or morphosemantic feature value's distribution is in the inflectional system of a language, the less a word-form expressing that feature value is expected.
This means that from the perspective of the morphological system, there are two logically distinct ways in which a word-form might “go missing.” One possibility is that there is an obviously well-defined paradigm cell, but no word-form to fill it; this is easily identified as inflectional defectiveness. But another possibility is that the paradigm cell itself is weakly motivated as a result of one or more of the relevant feature values being poorly integrated into the inflectional system of the language. This does not by itself constitute inflectional defectiveness. If there is a lack of (distinct) morphological marking but no unmet syntactic need, this is syncretism, or something conceptually close to syncretism (e.g., a subvalue). (See Chapter 4 for discussion.)
Where it gets interesting is when these issues overlap – examples in which a morphosyntactic feature value is only weakly motivated for some lexeme, but to the extent that it is, there is no word-form to express it. We can therefore formulate the same kind of loose scale along which instances of inflectional defectiveness can be ranked. The weaker the expectation of a word-form to express some value or set of values, the more difficult it is to know whether we are looking at defectiveness. Or more precisely, I argue that such a scenario is still an example of inflectional defectiveness, but less canonically so, exactly because the expectation for a word-form is less when the motivation for the corresponding morphosyntactic feature values is also less.
2.8.2 Weak defectiveness
Examples of gaps that are weak or marginal because they occur in the context of weakly integrated morphosyntactic values are fairly widespread and reflect varying degrees of strength of expectation. Space does not permit documentation of the full range of these examples, but Eastern Pomo, an agglutinative language of California, USA (Hokan family), can be used to illustrate defectiveness in the context of weak expectations. It has a pattern of marginal defectiveness that is in many respects the defective correspondent of limited expression of case and definiteness in Albanian. The following presentation of the case system is based on McClendon (Reference McClendon1977).Footnote 20 The discussion is limited to morphosyntactic cases that appear in the ninth position class slot.Footnote 21
In Eastern Pomo, all non-verbal open lexical classes encode case: pronouns, kinship terms, demonstratives, inanimate nouns, numerals, adjectives, and personal nouns. For these, nine morphosyntactic cases are expressed in the ninth position class slot: subject, direct object, indirect object, genitive, comitative-reciprocal, second allative, purposive, desiderative-abessive, and vocative. However, any given lexeme expresses only a subset of these nine, depending on which lexical class it belongs to. For instance, pronouns inflect for seven of the nine cases, but not desiderative-abessive or vocative.Footnote 22
(17) Pronouns (1pl forms)
sbj wá: ‘we’ dir obj wá:l ‘us’ indr obj wá:l ‘us’ gen wáybax ‘our, of us’ com-recp wáyMak̓ ‘with us’ second allFootnote 23 wáyNalal ‘to (in the direction of) us’ purp wáyba ‘for us’ desd-abes – voc –
In contrast, kinship terms have both desiderative-abessive and sometimes distinct vocative forms, but they lack second allative forms (‘motion towards’).Footnote 24
(18) Kinship termsFootnote 25
sbj ká:c̓ ‘my mother's father’ dir obj ká:̓čal ‘my mother's father’ indr obj ká:̓čiNàl ‘my mother's father’ gen hárik̓ibax ‘my father's’ hárik̓ ‘(my) father’ com-recp dá:riMak̓ ‘with (his) wife’ dá:ṭ̓ ‘(his) wife’ second all – purp dá:riba ‘for (his) wife’ desd-abes dá:riko ‘wishing for (his) wife’ voc hárik̓a ‘father!’
Inanimate nouns, like pronouns, lack a vocative form. However, they express both the second allative and the desiderative-abessive.
(19) Inanimate nouns
sbj xó ‘fire’ dir obj xó ‘fire’ indr obj xó ‘fire’ gen xóbax ‘fire's, of the fire’ com-recp ná:wišMak̓ ‘with the skirt, skirt and all’ ná:wiš ‘skirt.sbj/obj’ second all káNalal ‘to (his own) house’ ká ‘house.sbj/obj’ purp šá:riba ‘for (the) basket’ šá:ri ‘basket.sbj/obj’ desd-abes xóyi:kò: ‘for the sake of fire’ voc –
Personal nouns, demonstratives, adjectives, and numerals express yet other combinations of cases. Thus, the picture is that non-verbal lexemes in Eastern Pomo have word-forms for many of the same case categories, but not exactly the same ones. The question is are Eastern Pomo pronouns defective for desiderative-abessive and vocative? Are kinship terms defective for the second allative? Are inanimate nouns defective for vocative? And so on. These are not easy questions to answer.
First, notice that the classes share formatives. For instance, the comitative-reciprocal is realized by -Mak̓ for all lexical types. There are some differences of marking, but these reflect either phonological alternations or marking for animacy (e.g., -al for direct object marking on kinship terms versus zero marking on inanimate nouns). In other words, all non-verbal lexical classes in Eastern Pomo are subject to the same set of inflectional realizational rules, and when some case values are unexpressed for a given lexical class, it can seem obvious that we are looking at defectiveness (Aristar Reference Aristar1997: 347).
However, we should be cautious in this conclusion exactly because all pronouns fail to express desiderative-abessive and vocative, all kinship terms fail to expressive allative, and all inanimate nouns fail to express vocative. In this respect Eastern Pomo is quite different from Russian, in which verbs like pobeditˈ are easily identified as defective because the 1sg paradigm cell in the vast majority of verbs belonging to the same morphological class have well-formed first person singular forms. Stated simply, the expectation of a 1sg cell for pobeditˈ is clearly motivated by the behavior of other lexemes in the same part of speech. In Eastern Pomo this is not true.
The fundamental issue is thus whether the “missing” cases are motivated in Eastern Pomo. McClendon differentiates animates (pronouns and kinship terms) from substantives (inanimate nouns, demonstratives, numerals, adjectives, and personal nouns) based on morphological differences (e.g., substantives inflect for nominal aspect but animates do not), but it is unclear what the justification is for treating pronouns and kinship terms as belonging to different lexical classes, or for differentiation within the class of substantives. The available data are thus less complete than might be desired. Nonetheless, the fact that both animates and substantives inflect for case suggests that at the very least, there is some degree of expectation that all of these word classes should express the full set of nine case values. Moreover, given that all of these word classes inflect using the same inflectional formatives, in the same position classes, we might accept this as additional evidence that parallel behavior is expected. To this extent, the paradigm cells that lack any morphological expression (e.g. desiderative-abessive for pronouns) are defective but are much more marginal examples of defectiveness in comparison with examples like Russian pobeditʹ.
2.8.3 Canonical and non-canonical defectiveness
Baerman and Corbett (Reference Baerman and Corbett2010: 2) employ the idea of canonical defectiveness to describe this kind of variability: “the more idiosyncratic and lexically restricted the gap, the more canonically defective it is.” In the context of the formal definition of inflectional defectiveness in section 2.3, we can make this intuition more precise.
I propose that the right way to think about canonical or non-canonical defectiveness is in terms of the degree to which expectations are violated. Russian verbs like pobeditʹ are notable examples of defectiveness because the paradigm cell is unambiguously well defined. Expectations that there be a 1sg non-past form are thus high, and the lack of such a form is a large violation of expectations. In Eastern Pomo, however, expectations are lower to begin with, largely because entire lexical classes fail to express the relevant cases.
Earlier in this chapter, defectiveness was defined in the context of the morphosyntactic expression requirement and the syntactic need requirement. While this definition can be taken as a set of minimum requirements for defectiveness, we can identify some examples of paradigmatic gaps as more canonical than others by using the same criteria and specifying maximal requirements, rather than minimal ones.
(20) Canonical defectiveness occurs in the context of maximal morphological expression and maximal syntactic need.
Maximal morphological expression occurs when a value exhibits both unlimited applicability and unlimited combinability. Syntactic need is greater when the relevant value has a wider domain, and is lesser when it has only a narrow domain. Russian pobeditʹ is more canonically defective than Eastern Pomo pronouns are because morphological expression of case, number, and tense is more canonical in Russian than morphological expression of case is in Eastern Pomo, and there is greater syntactic need for the missing values in Russian than in Eastern Pomo.
The Eastern Pomo examples still fall under the umbrella of the term inflectional defectiveness, albeit as a non-canonical instance of the phenomenon. Ultimately, whether we attach the label “defective” to some data is not as important as the observation that the absence of some word-forms is a greater violation of expectations than the absence of others.
2.9 The (in)significance of (not so) isolated attestations
A final issue has to do with attestations and, more specifically, with how we evaluate the existence of forms that “fill” supposedly defective paradigm cells. The existence of a gap in some lexeme (or even the existence of defectiveness as a phenomenon more generally) might be challenged on the grounds that in the era of the Internet, it is possible to find examples of virtually any possible word in any widely used language. After all, speakers are wont to innovate and variation is rampant.
So what conclusions should be drawn when word-forms that correspond to supposedly defective paradigm cells are attested? The following discussion uses the English example of forgo to outline an interpretation of paradigmatic gaps as relative non-use. It is meant only to give the flavor of the argument, but the line of argumentation is nothing more than basic statistical reasoning and should be familiar to most readers. Chapter 7 implements the principle for Russian in a more rigorous manner.
The verb forgo (spelled alternatively forego) is commonly considered to have a paradigmatic gap in the past tense (e.g. Frampton Reference Frampton, Andronis, Ball, Elston and Neuvel2001). Relevant to the existence of the gap is the fact that there are two forms, for(e)went and for(e)goed, that directly compete. Many English speakers seem to have some uncertainty regarding the “correct” form. The following examples were returned by a Google search:Footnote 26
-
a. I for-what? I want the past tense of “forgo.” I forwent? I forgone? I forleft? Anyway, I avoided reading the bio of Zelda Fitzgerald.
b. I got it installed, but I foregoed (forewent?) the IIS install for now.
c. Because of the fussiness of the decals, I forgoed… forewent … err … I didn't put on any of the small stencil decals.
To many people, neither form sounds natural (but foregoed typically sounds worse), so on the rare occasion that the past tense is needed, avoidance may be preferable to choosing between bad options. But note that undergo, which is structurally identical to forgo, is entirely unproblematic (underwent / *undergoed). Thus, it is clear that the past tense gap in forgo cannot be explained entirely by the existence of both a regular and an irregular pattern. The much greater frequency of undergo compared to forgo presumably plays a role here, and perhaps other issues do as well.
The question that concerns us here is the consistency with which speakers circumlocute. The British National Corpus (BNC) contains 284 instances of the lexeme forgo, but not a single example of forgoed, foregoed, forwent, or forewent.Footnote 27 A token count of 0 is certainly consistent with the idea that the past tense of forego has a gap, but not definitive proof of it. However, a larger sample consisting of pages returned by a Google search produced about 229,400 page hits for forewent and forwent combined, and 157,200 page hits for foregoed and forgoed.Footnote 28 Does this mean that forgo does not have a paradigmatic gap in the past tense?
I argue that the absolute number of attested examples of a particular inflected form is basically irrelevant to the question of whether the paradigm cell in question is defective. It is the comparison of observed to expected frequency that is important. The number of page hits for foregoed/forgoed/forewent/forwent is in absolute terms a large number. However, the past tense constitutes only 0.2% of the total number of indexed pages returned by a search for forms of the lexeme forgo (=71.704 million hits). Given that the semantics of the verb are fully compatible with past tense, and that the past tense of verbs is frequently used in English overall, this is a significantly lower number of attestations than we might expect. By comparison, in both go and undergo, past tense forms represent slightly more than 20% of total lexeme frequency. Even if we (very conservatively) estimate that English verbs are, on average, used 5% of the time in the past tense form, we would expect to find about 3.5 million hits for the past tense of forgo – an order of magnitude more than the number of examples that we actually do find. In finding only 386,600 attestations, we can conclude that the number of observed forms is substantially lower than the number of expected forms and infer that speakers may often avoid using the past tense of forgo.Footnote 29 In this sense, a gap is a gradient statistical phenomenon.
A statistical perspective also highlights a crucial difference between rarely used but non-defective lexemes and lexemes with paradigmatic gaps. In the latter, the word-form representing the defective cell is used much less frequently than expected, given the overall frequency of the lexeme and the relative frequency with which that cell in the paradigm is used across lexemes. Infrequent usage is a deviation from expectations. By contrast, in a non-defective lexeme that is overall of low frequency, each word-form is not expected to be used often. Infrequent usage is thus in line with expectations. This further supports the idea that the important metric for identifying defectiveness is the number of attestations in the context of expectations, rather than the number of attestations in isolation.
2.10 Conclusions
This chapter offered a working definition of inflectional defectiveness and identified some challenges involved in trying to define the phenomenon empirically. The most important thing to emerge is the idea that paradigmatic gaps exist inherently at the interface between morphology and syntax. At one or the other level, they bear some similarity to phenomena that are ultimately distinct from defectiveness, but can be differentiated from them by looking at both levels of structure jointly.
The discussion in this chapter also highlighted the difficulty of defining inflectional defectiveness on purely empirical grounds. Since gaps fall at the intersection of morphology and syntax, fundamental theoretical assumptions about the nature of the morphology–syntax interface have a significant impact on issues such as how morphosyntactic feature structure is analyzed (and how any potential example of defectiveness is assessed against that analysis) and the relationship between defectiveness and periphrasis. In the end, a purely empirical definition of the phenomenon may be elusive.
Finally, since paradigmatic gaps are the lack of a word-form where we expected to find one, we must decide what to do about attested examples that “fill” a supposedly defective paradigm cell. Do such examples indicate that the lexeme is not, in fact, defective? I argued that attested examples should be interpreted in the context of expectations – a gap exists where there are far fewer attestations than would be expected to be found (assuming that the other conditions of the definition in (1) are met). This is partly a practical issue. However, it is also a theoretical issue: morphosyntactic features are integrated into inflectional systems to a greater or lesser degree, and syntactic structures make greater or lesser use of them. While the definition thus operates as a set of minimal requirements for defectiveness, within the set of examples that qualify as defective we can identify some as more canonical than others. Canonical defectiveness occurs in the context of maximal morphological expression and maximal syntactic need. Stated differently, defectiveness is inherently gradient because the expectation of finding a word-form to express a certain set of morphosyntactic values is itself gradient. Moreover, the same basic principle applies to attestations; it makes sense to think of paradigmatic gaps as gradient phenomena along this dimension as well. In terms of frequency of use, the relative lack of a word-form can be a greater or lesser deviation from expectations. This gradience is interesting by itself, inasmuch as it does not easily jive with the idea that defectiveness represents an aberration in inflectional systems, contrary to common assumptions.
Now that the definitional groundwork has been laid, we can move on to the main work at hand – beginning to explore what insight paradigmatic gaps offer into inflectional structure. In Chapter 3 I survey some of the types and causes of inflectional defectiveness. The data show that many languages have patterns of defectiveness that are fundamentally intertwined with those languages’ inflectional systems. This chafes even more clearly against the idea that patterns of defectiveness are (necessarily) superficial or random failures that can be accounted for separately from the core functioning of inflection. While many patterns of defectiveness are indeed lexically idiosyncratic, I argue that this does not contradict the idea that they are subject to and participate in general principles of inflectional organization.
