(Milne Reference Milne and Zaxoder1965: 75)
1.1 The problem of missing word-forms
A frequently cited truism about inflection is that it is highly productive. Inflectional morphology exists primarily to fulfill the needs of the syntax, which suggests that any given lexeme should have a form for every syntactic context in which it can appear, and this expectation is usually met. While speakers do not always have the same intuitions about what the form of a word should be, both children and adults readily produce required grammatical forms for both existing and novel lexemes (e.g., Albright and Hayes Reference Albright2002; Aronoff Reference Aronoff and Fisiak1980; Bauer Reference Bauer2001; Berko Reference Berko1958; Bybee and Slobin Reference Bybee and Slobin1982). Correspondingly, when a new lexeme comes into a language, it tends to be quickly integrated into the inflectional system, sprouting a full array of inflected forms. To take a single example, the verb google famously entered English in the late 1990s and developed the forms googles, googled, googling, will google, etc.Footnote 2 Examples of this kind, in which inflectional morphology applies seemingly automatically to new (and existing) lexemes, are numerous.
Table 1.1 An example of a defective Russian verb
| sprositˈ ‘to ask’ | singular | plural | pobeditˈ ‘to defeat’ | singular | plural |
|---|---|---|---|---|---|
| 1st person | sprošu | sprosim | 1st person | – | pobedim |
| 2nd person | sprosišˈ | sprosite | 2nd person | pobedišˈ | pobedite |
| 3rd person | sprosit | sprosjat | 3rd person | pobedit | pobedjat |
Against this backdrop, it seems surprising that in many if not all inflecting languages, we find some lexemes whose inflectional paradigms have Swiss-cheese holes in them – places where syntactically required forms simply fail to exist. Inflectional defectiveness can be defined intuitively as a situation in which one or more inflected forms of a lexeme is missing. It is the lack of any word-form when we expect to find some word-form. (A more formal definition is offered in §2.3.) Instances of inflectional defectiveness are paradigmatic gaps, or simply gaps for short.
Paradigmatic gaps are generally considered to be inflectional anomalies. A classic example of defectiveness – the pattern represented by the Russian verb pobeditˈ ‘conquer, defeat’ – serves to illustrate the basic issues.
The vast majority of Russian verbs express six combinations of person–number values in the non-past tense, including first person singular (e.g., sprositˈ ‘ask’). However, a number of verbs lack 1sg forms, including pobeditˈ. Zaliznjak (Reference Zaliznjak1977: 159) says that the first person singular of this verb is “difficult,”Footnote 3 and counts from the Russian National Corpus confirm that in the modern language, the expected 1sg form pobežu is not used to any observable degree, nor is any other conceivable 1sg form of this verb, despite the lexeme being fairly frequently used overall. The corpus contains 1,380 tokens of pobeditˈ in the other five non-past person–number combinations combined, but not a single token of pobežu as a verb form.Footnote 4 Russian speakers also deem any sentence containing the first person singular of this verb to be ungrammatical. Compare (1a), which is an internet newspaper headline,Footnote 5 to (1b), which speakers categorically reject, despite the fact that the sentence is semantically and syntactically unproblematic.Footnote 6
(1)
a. Kandidaty ot oppozicii pobedjat vo vsex pjati problemnyx okrugax. candidates from opposition win.pfv.npst.3pl in all five problematic regions ‘Opposition candidates will win in all five problematic regions.’ b. *Ja pobežu vo vsex pjati okrugax. I win.pfv.npst.1sg in all five regions ‘I will win in all five regions.’
The bottom line is thus that while a 1sg form of pobeditˈ is syntactically expected, there is simply no acceptable way to express the intended meaning, except by using a near synonym or other circumlocution – e.g. oderžu pobedu ‘I will score a victory’, in which pobedu is the accusative singular of the noun pobeda ‘victory’. Pobeditˈ has a paradigmatic gap in the 1sg non-past. If inflectional morphology is really so productive, and new lexemes acquire a full set of inflected forms automatically, then the failure of the first person singular is perplexing.
Moreover, the fact that only a small and lexically idiosyncratic group of verbs is affected makes the pattern of defectiveness all the more anomalous. Halle (Reference Halle1973: 7) states without elaboration that in Russian, “about 100 verbs” are defective in the same way that pobeditˈ is. Alley et al. (Reference Alley, Sims and Brookes2006) identify ninety-six lexemes, some based on the same root, that are listed as having 1sg gaps in at least one of eight reference works consulted. These include the verbs in (2), which are cited as defective in at least five of the sources. While a few of these are relatively unlikely to be used in the first person singular (e.g., šelestetˈ, which describes the noise made by dry leaves), in other cases the lack of a 1sg form is equally as surprising as in pobeditˈ.
(2)
bdetˈ ‘keep watch’ buzitˈ ‘create scandal, disorder’ galdetˈ ‘make a lot of noise’ derzitˈ ‘be impertinent’ dudetˈ ‘play the fife’ erunditˈ ‘do or say s.t. stupid’ zatmitˈ ‘darken; overshadow’ kudesitˈ ‘do magic’ očutit'sja ‘find o.s.; come to be’ rysitˈ ‘trot’ soseditˈ ‘be a neighbor’ ubeditˈ ‘convince’ umiloserditˈ ‘take pity on’ čudesitˈ ‘do magic’ čuditˈ ‘behave eccentrically’ šelestetˈ ‘rustle’ škoditˈ ‘misbehave’
The defective lexemes all belong to the same morphological subclass, but they represent only a small percentage of all verbs in this class. Thus, although there are some regularities to the distributional pattern, it is impossible to identify a morpho(phono)logically defined class that includes all and only the defective lexemes.Footnote 7 Similarly idiosyncratic examples of defectiveness have been documented in a variety of languages’ inflectional systems. Such gaps are often noted as rare but troublesome exceptions to the productive nature of inflection.
It is important to note that paradigmatic gaps do not necessarily result in significant disruption to communication and speakers need not be consciously aware of the existence of a gap. For instance, as a native speaker of English, I did not realize that beware has a defective paradigm until it was pointed out to me (Fodor Reference Fodor1972). The fact that speakers are capable of navigating around paradigmatic gaps without being (consciously) aware of their existence is noteworthy. In this respect, paradigmatic gaps are a little like suppletion, heteroclisis, deponency, and so on, which have the potential to be stumbling blocks, but which speakers navigate successfully. (It is mostly from the perspective of the linguist that these kinds of form–meaning mismatches seem remarkable.)
At least sometimes, however, gaps are indeed disruptive, creating impediments to communication. In everyday conversation, speakers start sentences and then cannot finish them because a required word-form does not exist. Speakers may also try out one or more possible forms before rejecting all of them (Albright Reference Albright, Garding and Tsujimura2003; Orgun and Sprouse Reference Orgun and Sprouse1999), as in the following example from the Internet, in which the writer runs into the past tense of forgo, which for many English speakers is defective.Footnote 8
(3) Because of the fussiness of the decals, I forgoed…forewent…err…I didn't put on any of the small stencil decals.
In such circumstances, there is no choice but to start the sentence over with a different verb or find some other work around.
Humor offers a less straightforward but interesting illustration of the disruption that gaps can cause. A charming example is found in Vinni-Pux i vse-vse-vse (Milne Reference Milne and Zaxoder1965), a Russian translation of the A. A. Milne story The world of Pooh, about Winnie-the-Pooh, a lovable but not very bright bear, and his friends in the Hundred Acre Wood. In one part, quoted at the beginning of this chapter and repeated below, Winnie-the-Pooh sings about his love for honey and how he will conquer heat and frost in order to get it. However, he keeps stumbling over the 1sg form of the verb pobeditˈ ‘conquer’.
(4)
Xorošo bytˈ medvedem, ura! It is good to be a bear, hurrah! Xorošo bytˈ medvedem, ura! It is good to be a bear, hurrah! Pobežu… I will conquer (pobežu)… (Net, pobedju!) (No, I will conquer (pobedju)!) Pobedju ja i žaru i moroz, I will conquer (pobedju) both heat and frost, Lišˈ by mёdom byl vymazan nos! As long as my nose is smeared with honey! Pobedju… I will conquer (pobedju)… (Net, pobeždu!) (No, I will conquer (pobeždu)!) Pobeždu, ja ljublju bedu, I will conquer (pobeždu), I love trouble, Lišˈ by byli vse lapki v medu!… As long as all my paws are in honey! (Milne Reference Milne and Zaxoder1965: 75)
The song plays with the fact that Russian verbs are expected to express six combinations of person–number values, yet pobeditˈ lacks a 1sg form. Winnie-the-Pooh tries out three different possible forms – pobežu, pobedju, and finally, pobeždu – but he simply cannot hit on an acceptable one. The humor thus lies in the fact that the structure of the Russian language makes it impossible for the sometimes-hapless Winnie-the-Pooh to conquer (!) the verb; his attempt to navigate the language is ill-fated from the start. While this playful use of language is obviously intentional, it illustrates that even from the perspective of speakers, a form is expected but missing.Footnote 9
Pobeditˈ thus serves to highlight the fundamental issue surrounding inflectional defectiveness: Inasmuch as gaps are disruptive to a language's morphosyntactic system, we should wonder why they arise and why speakers do not quickly and automatically generate word-forms to fill the empty paradigm cells, especially since gaps occur in only a small and idiosyncratic group of verbs. We might expect paradigmatic gaps to be fleeting phenomena – rare to occur and quickly smoothed out when they do surface. Yet they arise occasionally, and when they do, they can persist indefinitely. Many of the Russian 1sg gaps have persisted for over a hundred years (Baerman Reference Baerman2008), and they continue to be passed on to and learned by new generations of speakers. Paradigmatic gaps are thus fascinating primarily for the way that they seem to contradict the fundamental nature of inflectional morphology and fly in the face of speakers’ tendency to generalize. If for no other reason, this makes inflectional defectiveness an important topic for morphological study.
In this book, I explore what inflectional defectiveness can reveal about the structure of inflectional systems. Specifically, I am interested in the following questions:
(1) What mechanisms can account for the fact that inflectional morphology applies robustly to new lexemes, and also the fact that paradigmatic gaps arise and persist? Are these contradictory outcomes and how are they balanced against each other?
(2) What conditions lead to defectiveness, and why do speakers sometimes choose to leave a paradigm cell empty, rather than applying some kind of repair strategy?
(3) Once gaps arise, how are they learned by new generations? In particular, why do gaps often persist in a language, even in the absence of the conditions that originally gave rise to them?
(4) How should paradigmatic gaps be represented within a theory of inflection? Are paradigmatic gaps inflectional anomalies? Should defectiveness be accounted for outside of the normal functioning of inflectional structure or in a way that is integrated with it?
I argue that the answers to these questions lead to deeper insights about inflectional structure. One of the best ways to understand how linguistic systems work is by looking at how they break.
1.2 Random anomalies, epiphenomena, or (almost) normal morphological objects?
A useful place to start is by looking at how previous research has answered this last question. Should defectiveness be accounted for outside of the normal functioning of inflectional structure, or in a way that is integrated with it? In essence, this asks how much we should expect to learn about inflectional structure from studying defectiveness. Three broad positions – what I call the “gaps-as-anomalies” position, the “gaps-as-epiphenomena” position, and the “gaps-as-morphological-object” position – are identified and illustrated in this section. Note that the analyses discussed here should not necessarily be taken as definitive; the data under investigation have all been subject to alternative analysis. But each reflects a general philosophical perspective on the relationship between defectiveness and inflectional structure that has some degree of currency in linguistics.
The tension between the tendency of lexemes to have complete paradigms on the one hand and the existence of paradigmatic gaps on the other has long been recognized, but defectiveness has not until recently been a topic of much theoretical investigation. Prior to the early 2000s, if paradigmatic gaps were discussed at all, they tended to be treated as random anomalies. In an early account that looked at the Russian 1sg gaps, Halle (Reference Halle1973) posited that the verb form corresponding to the 1sg cell is generated according to productive inflectional rules, but in defective lexemes, a lexically specified feature [−Lexical Insertion] acts as a filter, preventing the generated form from being inserted into syntactic structure. The implication was that the Russian gaps are accidental exceptions to the normal functioning of the inflectional system. He labeled the Russian gaps as “arbitrary.”
More recently, Orgun and Sprouse (Reference Orgun and Sprouse1999) put forward a substantially similar analysis of inflectional gaps in Turkish and Tiene that they frame within Optimality Theory. They argue for a constraint evaluation component, CONTROL, which applies after EVAL. Within EVAL, candidate forms are evaluated according to the principles of classical Optimality Theory: there is competition between candidates to identify the one that best satisfies ranked constraints and constraints are violable if superseded by a higher-ranked constraint. Within CONTROL, however, only the winning candidate from the EVAL component is evaluated and constraints are never violable, so a candidate form which violates any CONTROL constraint results in a gap.
Orgun and Sprouse's formalization takes advantage of a general shift in phonological theory towards evaluation of surface well-formedness, and in this respect their analysis has a more general theoretical basis than Halle's does. However, like the feature [–Lexical Insertion], the CONTROL component serves a single purpose – to prevent productively generated forms from being inserted into syntactic structure. And like Halle, Orgun and Sprouse locate the formal account of defectiveness outside of the productive grammar. Fanselow and Féry (Reference Fanselow, Féry, Fanselow and Féry2002) expand on arguments for the CONTROL component and make an even clearer claim that defectiveness should be accounted for outside of the functioning of “core” grammatical mechanisms.
There is very little about this kind of approach to suggest that gaps interact in meaningful and interesting ways with other aspects of grammar, or that they even have the potential to do so. Since linguists are generally interested in productive grammatical phenomena, it would be easy to assume that inflectional defectiveness is not a rich topic for investigation and that we can simply relegate paradigmatic gaps to the status of idiosyncratic exceptions and move on to investigating more productive areas of the grammar. It is unsurprising that with the exception of Hetzron (Reference Hetzron1975), Halle's paper inspired few theoretical studies of inflectional defectiveness.
However, more recent research has hinted that paradigmatic gaps do, in fact, interact with inflectional structure in interesting ways. One line of thought has argued that defectiveness is an epiphenomenal result of morphophonological or morphosyntactic competition. For instance, Hudson (Reference Hudson2000) explores the English negative 1sg present gap in the copula that is represented by *amn't. He argues that the lack of a form for the negative 1sg cell is rooted in the organization of inflectional structure in terms of a multiple default inheritance hierarchy. In an inheritance hierarchy, general information that is shared by different elements is specified higher in the tree and then inherited by default by lower nodes, capturing the shared quality. The information structure thus becomes more specific as you move down the tree. Essentially, Hudson argues that the morphosyntactic feature specification of the negative 1sg present subsumes that of both aren't 〈be, {present, negative}〉 and am 〈be, {present, 1st, singular}〉, and so it should inherit its form from both. However, since the stem shapes are- and am- are in conflict, multiple non-orthogonal inheritance causes an unresolvable conflict, and thus, a gap. The important issue here is that the amn't gap is not lexically specified. In fact, there is no direct correspondent to the gap in Hudson's account. It is an epiphenomenal by-product of the conflicting morphosyntactic feature specifications.Footnote 10
Similarly in some respects, but operating more directly at the level of form, Albright (Reference Albright, Garding and Tsujimura2003) derives two patterns of defectiveness among Spanish verbs from morphophonological rule competition. Both patterns of gaps are connected to stem alternations; for instance, defectiveness occurs in exactly the inflection classes and present tense paradigm cells that are (potentially) affected by diphthongization or raising of the stem vowel under stress – i.e., in all present tense singular forms and third person plural, but not first or second persons plural. (This “N-pattern allomorphy” [Maiden Reference Maiden2004] is observed in many Romance languages and it is tied to defectiveness in several.) Based on the principles of the Minimal Generalization Learner (Albright and Hayes Reference Albright2002), Albright proposes that low lexeme frequency and low reliability of the alternation combine to produce gradient uncertainty within the inflectional system, the extreme result of which is a paradigmatic gap. In essence, he argues that the grammar is indeterminate because the affected lexemes fall into an area of the lexicon in which there is no island of reliability (high confidence implicative relation) from which to derive the form. This derives the distribution of paradigmatic gaps from the core functioning of the inflectional system. Crucially, however, it also treats them as epiphenomena and denies them any direct grammatical status.
Hudson argues that the *amn't gap – and by implication, all paradigmatic gaps – must fall out from conflicting grammar principles in this way because otherwise they would not be learnable.Footnote 11 Albright is neutral on the question of whether all gaps must fall out epiphenomenally from the grammar. However, he argues that lexical specification of the type proposed by Halle predicts that gaps will be randomly distributed in the lexicon. The implication is that gaps of the Spanish (and Russian 1sg) type cannot be adequately treated as lexically specified because their distributional facts indicate that they are not random accidents. He thus seems to make a sharp distinction between gaps that are grammatically motivated (and appropriately treated as epiphenomena) and gaps that are random accidents (and appropriately treated via lexical specification).
In short, previous analyses have often either treated gaps as the by-product of morphological rule competition or dismissed gaps as random accidents to be accounted for through ad hoc lexical specification. Either gaps are regulated by productive grammatical principles to the point of not requiring (and perhaps not even allowing!) direct generalization, or they are anomalous to that structure and lie entirely outside of it. These represent opposite approaches in many respects. However, in reducing gaps either to epiphenomena or to anomalies, proposals of the type discussed above share the assumption that defectiveness has a status that is fundamentally different from other kinds of morphological patterns.
This assumption has been challenged, however. For instance, Anderson (Reference Anderson2010) investigates the Surmiran Romansch (Indo-European > Romance, spoken in Switzerland) verb dueir ‘should’, several forms of which are missing. As in Spanish, the key generalization relates to the N-pattern distribution of stems. Some classes of Surmiran Romansch verbs have different stem shapes when stress falls on the stem (in present tense, all singular forms and 3pl) than when it falls on the desinence (1pl and 2pl). Defectiveness in dueir follows the same distribution: gaps exist exactly and only in those paradigm cells where we would expect the stressed stem allomorph. Moreover, interestingly, the verbs that are close phonological neighbors to dueir (e.g., stueir ‘must, should’) are suppletive. The core of Anderson's argument is that dueir is defective precisely because one of its required stems is not lexicalized and the resources of the language do not offer an obvious model on which to form the stressed stem allomorph, or at least, not one that speakers prefer to simply using a different verb.
In some respects, Anderson's argument has features in common with Albright's analysis, even beyond the fact that they both investigate variants of the Romance verb stem alternation pattern. In particular, Anderson's claim that there is no valid model on which a stressed stem of dueir could be formed requires that such generalizations are made on a very local scale. This has some overlap with Albright's notion of an island of reliability – both argue that gaps arise out of localized failures in the generalizability (productivity) of inflectional form. However, a crucial part of Anderson's analysis is the claim that the alternation in the verb stems is morphologized and that the stem alternants are therefore morphomic in the sense of Aronoff (Reference Aronoff1994).Footnote 12 The structure of the stem space is thus an element in the pattern of defectiveness, independently of the issue of grammar competition/indeterminacy.
Boyé and Cabredo Hofherr (Reference Boyé and Hofherr2010) argue even more directly that defectiveness in French and Spanish verbs (including some of the same Spanish facts considered by Albright) is sensitive to the organization of stem space, meaning, to the structure of the lexicon. They maintain that the gaps in these languages must be considered to be lexically specified; however, they draw parallels to stem suppletion. Stem suppletion has been shown in Romance verbs to follow the distribution of regular patterns of morphological stem alternation, suggesting that both regular and suppletive morphological patterns are governed (in part) by the same principles of stem organization, despite suppletion being in other respects highly lexically idiosyncratic (Aski Reference Aski1995; Bonami and Boyé Reference Bonami, Boyé, van Eynde, Hellan and Beermann2002; Boyé and Cabredo Hofherr Reference Boyé and Hofherr2006; Hippisley et al. Reference Hippisley, Chumakina, Corbett and Brown2004). They suggest that like suppletion, the patterns of verbal defectiveness in Spanish and French are both subject to general principles of inflectional organization, and at the same time are lexically idiosyncratic. They thus implicitly reject the idea that paradigmatic gaps (necessarily) have a fundamentally different status from other kinds of morphological patterns. Baerman and Corbett (Reference Baerman and Corbett2010) propose that lexicalized defectiveness should be conceptualized as a “morphological object,” i.e., as a generalization about the relationship between form and meaning, in some ways no different from any other such generalization.
Inflectional defectiveness is a diverse phenomenon and different kinds of account are likely to be appropriate to different instances of it. However, as the discussion above illustrates, the philosophical positions from which inflectional defectiveness has been approached are not all compatible. A fundamental issue that has threaded through the literature has been the question of whether defectiveness is anomalous to the normal functioning of inflectional structure. While “gaps-as-epiphenomena” accounts have sought to explain at least some paradigmatic gaps as the by-product of normal inflectional structure, both this general approach and the “gaps-as-anomalies” approach have equated lexically specified defectiveness with accidental exceptions to productive inflection. But others have rejected this equation and suggested a parallelism between (lexically specified) defectiveness and other aspects of inflectional structure. This latter perspective suggests that lexicalized defectiveness is a much richer topic for investigation than either the “gaps-as-anomalies” viewpoint or the “gaps-as-epiphenomena” viewpoint would imply.
In this book I put forward an argument that is in keeping with the philosophical stance of the “gaps-as-morphological-object” approach but takes a somewhat bolder position. Specifically, I argue that inflectional defectiveness is a systemic variant of normal inflectional structure.
1.3 Words, paradigms, and the organization of the lexicon
The argument that I develop is rooted in the dynamic, paradigmatic structure of the lexicon. I see this organization as crucial to understanding how defective and non-defective inflectional structures emerge and are organized. It is thus necessary here to lay some groundwork by saying a few words about the structure of the lexicon. The goal is to give only a broad overview of fundamental assumptions and the kinds of arguments that have been adduced in favor of them. There is not space here to fully justify positions or to replicate arguments that have been made elsewhere. Nor is there space to elaborate a fully precise model of lexical structure. However, the reader is referred to the cited works, and relevant elements of the model will be developed as the book progresses.
1.3.1 The lexicon is not like a prison
At least since Bloomfield (Reference Bloomfield1933), theories of language architecture have often relegated idiosyncratic morphological patterns to “the lexicon,” i.e., to a domain outside of the productive rules of grammar. Di Sciullo and Williams (Reference di Sciullo and Williams1987: 3) famously suggested that the lexicon “is incredibly boring by its very nature…Those objects that it does contain are there because they fail to conform to interesting laws. The lexicon is like a prison – it contains only the lawless, and the only thing its inmates have in common is lawlessness.” This draws a sharp distinction between that which is regular and rule governed (“law-abiding society”) and that which is idiosyncratic and irregular. Only the latter fall within the domain of the lexicon.
Ironically, a consequence of this architecture is that it makes rule-governed morphological structure incredibly boring. When morphology develops diachronically out of phonology or syntax, a frequent hallmark of this change is the splintering of a single, broad generalization into a series of more fragmentary and morpholexically conditioned generalizations (Joseph Reference Joseph, Joseph and Janda2003; Joseph and Janda Reference Joseph, Janda, Hammond and Noonan1988). Yet shifting the resulting synchronic “idiosyncrasies” to the lexicon has the effect of bleeding the synchronic morphological system of its richness. It leaves little other than morpheme combinatorics to be accounted for as part of rule-governed morphological structure. Not surprisingly, then, a sharp distinction between the idiosyncratic lexicon and the regular morphology is often concomitant of a view of morphology as the syntax of words. Consider, for example, Boeckx's (Reference Boeckx2008: 63–64) statement that “Words don't exist, they are the phlogiston of linguistics. Everyone must say that sentences are built out of morphemes…[We] expect a high degree of isomorphism, of the type expressed by Baker's (Reference Baker1985) Mirror Generalization; although various affixal properties lead to readjustment rules that end up masking syntactic structures.” In other words, “law-abiding” morphology is isomorphic with syntactic structure in a way that is captured in the Structuralist adage “one meaning – one form.” While various readjustments must be allowed, morphology at its core adheres to the Agglutinative Ideal (Ackerman and Malouf Reference Ackerman and Malouf2011). To this we can add something so obvious as to not even need stating: “law-abiding” morphology is productive.
So from a certain perspective, morphological structure is boring because it amounts to little more than productive morpheme concatenation. And the lexicon is boring because it is merely an unstructured repository of exceptions. This seems to be an unpromising perch from which to try to make sense of paradigmatic gaps as anything more than random anomalies. Yet it is clear that (some instances of) gaps are not random and as noted above, I will argue that they are not even anomalies. So in order to make sense of inflectional defectiveness, we must approach the topic from a different perspective.
An alternative, fundamentally morphocentric view rejects the idea that the lexicon is merely a repository for idiosyncrasy, or that the lexicon and morphological structure can be so cleanly walled off from each other. Maiden (Reference Maiden, Booij and van Marle2005: 168) phrases the issue as follows: “autonomously morphological structure need not be an inert, defunct residue of an earlier état de langue, nor a kind of diachronic ‘dead end’. It can be a dynamic, pervasive, self-reinforcing factor.” He frames this comment in terms of morphological change, but explicitly with the goal of motivating such structures in grammar architecture synchronically. Joseph and Janda (Reference Joseph, Janda, Hammond and Noonan1988: 204) state the issue even more bluntly: “Morphology is not a graveyard.” The implication is that even highly idiosyncratic phenomena are not by definition wholly anomalous to the functioning of a language's grammatical system.
Moreover, there is good reason to stage a jailbreak, which is to say, to posit a richly structured lexicon from which principles of morphological organization emerge. Despite Boeckx's exhortation that “everyone must say that sentences are built out of morphemes,” formal and cognitive/psycholinguistic lines of research have independently converged on the position that words (or even larger constructional units; see Booij Reference Booij2010) are the basic unit of the lexicon and of morphological structure, that morphemes have at best a secondary status, and that lexical elements are organized into a structured network of paradigmatically oriented connections.
1.3.2 Words and paradigms in formal morphology
Hockett (Reference Hockett1954) identifies three types of models of inflectional structure. In the paper he is chiefly concerned with comparing two of these: Item-and-Arrangement (IA) and Item-and-Process (IP) models. However, as an addendum apparently inserted after the rest of the paper had been written, Hockett mentions a third kind of model – Word-and-Paradigm (WP). The WP approach to inflection was subsequently elaborated by others (most notably, Anderson Reference Anderson1992; Aronoff Reference Aronoff1994; Matthews Reference Matthews1972; Robins Reference Robins2001[1959]; Stump Reference Stump2001; Zwicky Reference Zwicky, Niepokuj, VanClay, Nikiforidou and Feder1985).
These three types of models divide along two dimensions labeled by Stump (Reference Stump2001) as lexical vs. inferential and incremental vs. realizational. In a lexical theory, inflectional morphs are lexical entries and words are built by concatenating morphemes together. By contrast, in inferential theories, words are constituted by processes that apply to produce one form out of another. These processes may be concatenative or may involve other kinds of operations, including non-concatenative changes to a stem. Along the other dimension, an incremental theory posits that morphosyntactic values are added to a word as a consequence of morphological form being added to a root. Formal operations are thus meaning-adding. By contrast, a realizational theory posits that the combination of stem and a set of morphosyntactic values licenses the addition of morphological form. In this sense, formal operations are not meaning-adding and the meaning of a word is not expected to be the sum of the meanings associated with its formal operations. Both IA and IP model are incremental, in being meaning-adding, and in this sense they are both morphemic. The morpheme (whether a lexical entry or a rule) is the basic unit at which form and meaning are associated. They differ along the other dimension, however, in that IA models are lexical, whereas IP models are inferential. IA modes thus represent a more “classically morphemic” model of inflection. By contrast, WP models are inferential-realizational. In WP models, the basic level at which form and meaning are related is the word, not the morpheme. The morpheme is an epiphenomenon or, at most, a secondary kind of representation in the lexicon that is derived by generalizing over full word-form exemplars.
Several kinds of arguments have been put forward in favor of inferential-realizational models of inflectional structure; see Chapter 1 of Stump (Reference Stump2001) for an overview. One line of argumentation, stemming particularly from Matthews (Reference Matthews1972), focuses on the fact that inflectional structure is replete with form–meaning mismatches: form with no corresponding meaning (empty morphs), meaning with no corresponding form (zero expression), more than one meaning corresponding to a single form (cumulative expression), more than one form for a single meaning (extended exponence), a form which in one inflection class bears the opposite value of some morphosyntactic feature as in another class (deponency), inflected forms that seem to be built on already-inflected stems without carrying the latter's morphosyntactic values (parasitic/Priscianic formation), and so on. The issue here is that if form and meaning are associated at the level of the morpheme and words acquire morphosyntactic values via the addition of morphemes (items or processes), form–meaning mismatches are problematic, in that they result in underdetermination, overdetermination, or misdetermination of the morphosyntactic values of a word. Such phenomena thus challenge the notion that morphemes are lexical bundles of form and meaning, and the fundamental level at which form and meaning are associated. The principles of an incremental theory of inflection can be saved only by resorting to unmotivated and extraordinary mechanisms, like zero morphemes, that have the effect of rendering the theory unfalsifiable (Stump Reference Stump2001) and that exist for the sole purpose of saving the assumptions of incrementalism (Haspelmath and Sims Reference Haspelmath and Sims2010). Inferential-realizational approaches resolve this problem by positing an indirect relationship between form and meaning that gives primary importance to the level of the word.Footnote 13
Another line of argument in favor of words as basic units focuses on the need in IA models for an unmotivated distinction between concatenative and non-concatenative inflection (Zwicky Reference Zwicky and Levine1992). To the extent that morpheme concatenation is the most prevalent kind of morphological pattern in the world's languages, it can be argued that there is justification for according a special status to concatenation. However, it has been argued that the diachronic sources of morphological patterns are more likely to lead to concatenative morphological patterns than non-concatenative ones (Bybee and Newman Reference Bybee and Newman1995; Janda Reference Janda1982, Reference Janda1984) and that, more generally, the typological distribution of morphological structures often has a historical explanation (Harris Reference Harris and Good2008). There is thus no need to build a preference for concatenation directly into synchronic morphological theory.
A final argument has to do with poor evidence for branching hierarchical structure word-internally, especially in inflection (Anderson Reference Anderson1992; Janda Reference Janda, Richardson, Marks and Chukerman1983), suggesting that words do not have internal structure in the same sense that syntactic structures do.
As the name Word-and-Paradigm suggests, in this approach paradigms also become morphological objects of theoretical interest. Arguments have focused on the need for paradigmatically oriented relations to account for phenomena such as syncretism (Baerman et al. Reference Baerman, Brown and Corbett2005; Blevins Reference Blevins1995; Zwicky Reference Zwicky, Niepokuj, VanClay, Nikiforidou and Feder1985). While syncretism in particular has been a hot topic of debate between paradigmatic and non-paradigmatic approaches to inflection (compare the preceding references to Bachrach and Nevins Reference Bachrach, Nevins, Bachrach and Nevins2008; Bobaljik Reference Bobaljik2002; Müller Reference Maiden2011; Noyer Reference Noyer, Lapointe, Brentari and Farrell1998), overall, the argument in favor of a theory of inflection that incorporates direct paradigmatic connections between inflected forms is overwhelming. A second line of argumentation relates to inflection class structure – in particular, the co-implicative relations holding between inflected forms in the same class. Recent work has emphasized these implicative relations as the fundamental principle underlying inflection class organization (Ackerman et al. Reference Ackerman, Blevins, Malouf, Blevins and Blevins2009; Ackerman and Malouf Reference Ackerman and Malouf2013; Finkel and Stump Reference Finkel and Stump2007; Stump and Finkel Reference Stump and Finkel2013; Wurzel Reference Wurzel1989).
1.3.3 Words and associative networks in lexical representations and processing
Independently and in parallel to the formal morphological literature, psycholinguistic investigations of the lexicon and lexical access have similarly established strong evidence for words as representational units, and the need for an associative (i.e., paradigmatically oriented) network of connections between lexical entries. Early models of the lexical storage and processing of inflected forms focused on the extent to which inflected forms are decomposed during perceptual lexical access (Caramazza et al. Reference Caramazza, Laudanna and Romani1988; Prasada and Pinker Reference Prasada and Pinker1993; Taft Reference Taft1979; Taft and Forster Reference Taft and Forster1975), often coming to the conclusion that regular inflections are decomposed during lexical access while irregularly inflected forms are directly stored and accessed (Clahsen et al. Reference Clahsen, Eisenbeiss and Sonnenstuhl-Henning1997; Pinker Reference Pinker1991; Prasada and Pinker Reference Prasada and Pinker1993). However, such a strict division between regular and irregular inflection has proven untenable under the accumulated weight of evidence that regular inflected forms also often exhibit properties that are indicative of cognitive storage and access of the inflected form as a single unit.
One of the most robust and omnipresent results of studies of lexical processing is that lexical access speed is sensitive to word-form frequency. Robust sensitivity to word-form frequency has been found in perception (Burani and Thornton Reference Burani, Thornton, Baayen and Schreuder2003; Niswander et al. Reference Niswander, Pollatsek and Rayner2000; Raveh Reference Raveh2002; Taft Reference Taft1979) and production (Jescheniak and Levelt Reference Jescheniak and Levelt1994; Levelt et al. Reference Levelt, Roelofs and Meyer1999), and for both irregular and regular inflected forms (Alegre and Gordon Reference Alegre and Gordon1998; Butterworth Reference Butterworth and Butterworth1983; Katz et al. Reference Katz, Rexer and Lukatela2004). Other factors that have been shown to influence lexical access include the semantic transparency of a complex word (Feldman et al. Reference Feldman, Barac-Cikoja and Kostić2002; Schirmeier et al. Reference Schirmeier, Derwing and Libben2004), orthographic transparency during reading (Burani et al. Reference Burani, Barca and Ellis2006), affixal salience within a morphologically complex word (Järvikivi et al. Reference Järvikivi, Bertram and Niemi2006), morphological family size (De Jong et al. Reference De Jong, Schreuder and Baayen2000; Feldman and Pastizzo Reference Feldman, Pastizzo, Baayen and Schreuder2003), and affixal homonymy (Bertram et al. Reference Bertram, Laine, Baayen, Schreuder and Hyönä2000a; Bertram et al. Reference Bertram, Schreuder and Baayen2000b).
Taken collectively, the received picture is one in which lexical access is strongly sensitive to the distributional properties of word-forms. This requires that whole word-forms have representational status in the lexicon. The evidence that regular inflected forms are sometimes stored and accessed as whole words (e.g., Alegre and Gordon Reference Alegre and Gordon1998) is particularly important here, since it undercuts the idea of the lexicon as merely the repository of idiosyncrasies and irregularities.
Two classes of models – parallel dual-route (aka race) models of morphological processing (Baayen et al. Reference Baayen, Dijkstra and Schreuder1997; Baayen and Schreuder Reference Baayen and Schreuder1999) and parallel distributed processing (aka connectionist) models (Daugherty and Seidenberg Reference Daugherty, Seidenberg, Lima, Corrigan and Iverson1994; Elman et al. Reference Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett1996; papers in Rumelhart et al. Reference Rumelhart, McClelland, McClelland and Rumelhart1986) have approached the study of cognitive processing from opposite directions, and to different conclusions in important respects, but in both cases emphasizing the role of the influence that distributional properties of words have on lexical organization and processing.
At their most basic, parallel dual-route models claim that lexical processing is optimized for processing speed. They posit competition between parsing (aka computation) and direct access of morphologically complex words in the lexicon (aka storage). During lexical processing in perception, a complex word is processed simultaneously along two routes – via the access representation for the whole word and via the access representations for the word's (potential) morphological constituents. Whichever method is faster “wins the race,” which is to say that the word is accessed successfully via that method. Parallel dual-route models, more than some other models of lexical processing, thus reify morphemes, while seeing the balance between word-based access and morpheme-based access as highly gradient and sensitive to statistical-distributional properties that facilitate or inhibit access via one or the other route. However, even in this approach, morphemes do not preclude word-based representations, and in fact, secondary processing is generally assumed to increase the activation of both a word-form and its component morphological structures, regardless of initial access method. Morphemic lexical entries are thus generalizations over word-forms that exist side by side with word-form representations, not as a replacement for them. Plag and Baayen (Reference Plag and Baayen2009) suggest that all else being equal, there is an inherent bias toward word-form storage and access, at least in English. This not only offers a hint as to why words tend to pick up semantic idiosyncrasies over time but also offers further evidence for the fundamentally memory-rich nature of the lexicon.
Connectionist models, also known as parallel distributed processing models, differ from parallel dual-route models in a number of ways, most significantly in being subsymbolic. This entails a fundamentally different representational structure, in which the lexicon is conceptualized not as a list of entries akin to a dictionary, but as a series of neural nodes that are gradiently “on” or “off,” forming a pattern of activation. No single node (or group of nodes) is uniquely assigned to any particular language structure; the representation of a word, morpheme, sentence, etc. is instead “distributed” across the nodes of the system and captured in terms of its activation pattern. Conversely, any given node participates in the representation of many (and conceivably, all) words. It is thus false to talk about connectionist models as being either morpheme-based or word-based in any direct sense, since neither morphemes nor words as symbolic units have any exact correlate within a connectionist model. However, connectionist structure generalizes by exploiting regularities in the mapping between form and meaning. Thus, morphemes emerge as patterns of node activation to the extent that form–meaning correspondences are repeated in language input (e.g., words) and thus exploitable. The statistical-distributional properties of the input are the material from which morphological generalizations emerge. For the present purposes, the importance of this is that to the extent that the resulting generalizations correspond to structures that linguists identify as morphemes, these are emergent structures. In short, connectionist models leverage a memory-rich lexicon to capture gradient morphological structure.
Finally, a third line of thought, associated with the laboratory phonology tradition, has explored the paradigmatic dimension of lexical organization in particular. Of special interest here are lexical gang effects in morphological structure (Alegre and Gordon Reference Alegre and Gordon1999; Bergen Reference Bergen2004; Bybee and Moder Reference Bybee and Moder1983). A lexical gang is a set of words that are phonologically similar and share morphological behavior. For instance, the fifty or so English verbs with the deadjectival suffix -en (blacken, brighten, flatten, etc.) all have monosyllabic stems that begin with a consonant or consonant cluster and end in an obstruent. There is a preference for the obstruent to be a coronal and/or voiceless, although not all members of the gang conform to these properties. This group of words behave like a lexical gang in the sense that the boundaries of the gang are fuzzy rather than sharp, and speakers are gradiently sensitive to the phonological properties that define the gang and judge the well-formedness of nonce -en words according to how well they match the properties of “core” members of the gang (Alegre and Gordon Reference Alegre and Gordon1999). As Pierrehumbert (Reference Pierrehumbert, Cohn, Fougeron and Huffman2012: 178) notes, “Experimental results on lexical gangs and neighborhoods show that subsets of the full lexicon, defined as clusters of words that are particularly similar amongst themselves, have pervasive force. The results support a picture of the lexicon in which words are organized in a network, where the links represent shared phonological and semantic properties.” A central goal, then, is to understand how this rich network structure creates sub-networks, and how morphological structure emerges out of this network structure.
1.3.4 Convergences and almost-intersections
Formal inferential-realizational models of inflection and psycholinguistic studies of lexical access have operated almost exclusively in parallel, not in tandem. The objects of inquiry and research questions have been different in each. However, both are (directly or indirectly) interested in the structure of the lexicon, and in non-trivial ways the two literatures have arrived at some of the same conclusions. First, both assign a central role to word representations and minimize the significance of morpheme representations. Even in the psycholinguistic literature, where morpheme representations are sometimes given formal status, these can be construed as patterns that are extracted from word-level (or larger) representations and thus as having secondary status as units of structure.
Second, both lines identify associative/paradigmatic connections holding between inflected forms as the fundamental organizing principle of the lexicon/inflectional structure. In inferential-realizational models, paradigmatically oriented generalizations are seen as necessary for an empirically adequate description of inflectional structure, but at a more general level and more importantly, it is the opposition between elements (inflected words) in a system that constitutes and organizes morphological structure.Footnote 14 The individual elements bear no meaning and have no structure that is independent of the oppositions embodied in the inflectional system. Similarly in memory-rich models of the lexicon, such as those stemming from the laboratory phonology tradition, the network of associative connections defines the structural organization of phonology and morphology. These define the relatedness of elements, with neighbors (i.e., elements that are interconnected by virtue of sharing some element of form or meaning) exhibiting a pervasive influence on each other. These connections structure not only the lexicon but also morphological structure by being the basis for competition between alternative realizations.
At the same time, the fact that the formal inflectional literature and the psycholinguistic literature have pursued different questions means that they have contributed different insights. Partly as a result, there are many places where the two lines of work almost intersect – but not quite. For instance, the morphological literature has tended to focus on paradigms as theoretical constructs and formally bounded entities. By contrast, psycholinguistic studies of the lexicon have emphasized the paradigmatic dimension, but without a formal notion of the paradigm. When taken together, the two literatures raise interesting questions about the status of paradigms as reified entities, and to the extent that inflectional structure is characterized by paradigms, how this structure emerges. (I return to this issue in Chapter 8, in light of the intervening discussion of paradigmatic gaps.) The morphological studies have also tended to focus on paradigmatic relations among forms of the same lexeme (e.g., the paradigmatic dimension of syncretism), whereas as often as not, psycholinguistic models of the lexicon emphasize connections between different lexemes sharing some property (e.g., words belonging to the -en lexical gang). At a practical level, the notion of “paradigmatic” that is employed can be quite different. Finally, both have emphasized the importance of small-scale generalization, but in different ways. Formal models have tended to emphasize the principle that more specific generalizations preclude more general ones. They encode a formal notion of default realization rules. By contrast, in studies of the lexicon, locality is defined in terms of morphological and/or phonological neighborhoods and statistical reliability. No obvious translation exists from one to the other, despite being clearly related notions. In short, while each line of investigation has much to (potentially) offer the other, the points of connection are not as obvious as they might seem at first glance.
Ultimately, there is a remarkable if underappreciated convergence of ideas. Disparate lines of inquiry contribute different insights but offer interesting opportunities to understand phenomena that lie at their nexus. Some of these issues have already been widely recognized. For instance, the relationship between affix productivity (as an issue of morphological structure) and whole word lexical storage (as an issue of cognitive lexical organization) is well known (Aronoff and Anshen Reference Aronoff, Anshen, Spencer and Zwicky1998; Hay Reference Hay2003; Hay and Baayen Reference Hay and Baayen2002). In other areas, however, there is still much work to be done. In this book, I try to take advantage of a position at the nexus between formal morphology, linguistic typology, and psycholinguistics to shed light on the phenomenon of inflectional defectiveness.
1.4 Structure and organization of the book
The remainder of this book is organized as follows. Chapters 2 through 4 look empirically at the question of whether paradigmatic gaps are inflectional anomalies, and I make the case that in fact they are in many respects quite normal aspects of inflectional structure. Chapter 2 looks at definitional issues and the gradient line between defectiveness and well-formedness. Chapter 3 explores the causes of defectiveness based on a literature review. 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. Chapter 4 looks at the relationship between defectiveness and productivity. I argue that defectiveness has been mistakenly equated with a failure of productivity and that defectiveness and productivity are in fact orthogonal issues. This opens the door to asking how defectiveness competes with other kinds of morphological expression, in particular, syncretism.
Chapters 5 through 7 look in detail at two particular cases of inflectional defectiveness – gaps in the genitive plural of Modern Greek nouns (Chs. 5 and 6) and in the first person singular of Russian verbs (Ch. 7). Chapter 5 explores the role of paradigm cohesion, i.e., inflectional form inter-predictiveness, as a cause of gaps. Chapter 6 explores how defectiveness becomes irreducible (lexicalized). Chapter 7 asks how irreducible patterns of inflectional defectiveness are learned.
Finally, Chapter 8 sums up some general consequences/implications of this research for a general theory of paradigm structure.
