Reporting on a study of 79 languages (see appendix B), I argue that (morpho)syntactic structure plays a crucial role in two observed asymmetries: (a) in nouns, number-driven root suppletion is common while case-driven root suppletion is virtually unattested, and (b) in contrast to lexical nouns, pronouns commonly supplete for both number and case. I propose that the structural difference between lexical nouns and pronouns, combined with locality effects as proposed in Distributed Morphology (DM; Halle and Marantz 1993), accounts for the two asymmetries. This raises the question whether these asymmetries can be captured in frameworks that deny that hierarchical syntactic structure plays a role in the morphology, such as word-and-paradigm approaches (e.g., Anderson 1992, Stump 2001).
Suppletion refers to the situation where a single lexical item is associated with two phonologically unrelated forms, and the choice of form depends on the morphosyntactic context (see Corbett 2007 on specific criteria for canonical suppletion).1 Although rare in absolute terms, it is regularly observed across languages (Hippisley et al. 2004). For illustration, compare the (nonsuppletive) adjective-comparative-superlative paradigm smart-smarter-smartest, in which the root remains the same throughout, with the familiar example of good-better-best. In the latter case, the root in the adjective surfaces as good, but in the context of the comparative (and superlative) we see be(tt). With respect to nouns, languages can display suppletion for number (#). In Ket (Werner 1997), canonical nouns display a nasal suffix in the plural (1), but suppletive nouns display root suppletion in the same context (2).2
Curiously, though, as remarked briefly by Bybee (1985:93) and confirmed in the study of 79 languages described here, root suppletion of nouns in the context of case (K) seems to be largely unattested (apparent counterexamples are discussed in section 4.1).
In stark contrast, pronouns regularly display suppletion for number (see also Corbett 2005), as well as case. Consider, for instance, number-driven suppletion in 2nd person pronouns (3) and case-driven suppletion in 1st person pronouns (4) in Latvian (Mathaissen 1997).
Indeed, it is widely assumed that pronouns have less structure than lexical nouns (Postal 1969, Longobardi 1994, Déchaine and Wiltschko 2002). Crucially, I assume that lexical nouns (5) contain, at a minimum, a root and a category-defining node n. In essence, the n node will have the effect that the root and K are not sufficiently local.3 In contrast, pronouns (6) are, at their core, functional (D) (‘‘D’’ is used merely as a label) and thus lack the category-defining node that intervenes between the root and K in (5).4
In section 2, I first introduce the relevant background in which the analysis is couched, and briefly discuss the relevant notion of structural locality in morphology. In section 3, I discuss the structure of nominals in more detail and how it interacts with locality restrictions to prohibit case-driven root suppletion in lexical nouns. In section 4, I discuss some predictions that follow from the particular formalization of structural locality proposed here, showing further support for the role of locality in suppletion; and in section 5, I offer final remarks.
DM crucially incorporates hierarchical structure into the morphology; essentially, it assumes the input to morphology to be syntactic structure. Features (or feature bundles) are distributed over nodes, which in turn are subject to Vocabulary Insertion (VI). Furthermore, VI proceeds cyclically from the lowest element in the structure outward.
Suppletion is modeled as contextual allomorphy; that is, although a particular feature bundle has a corresponding exponent as a contextfree default, an exponent specified for a more specific context takes precedence (per the Elsewhere Principle; Kiparsky 1973). Consider again the good-better-best paradigm; its regular (context-free) exponent is good, but in the context of the comparative (and superlative) it corresponds to be(tt).
√GOOD ⇔ be(tt) / _____ comparative
√GOOD ⇔ good
Now, consider the representation of lexical nouns (see (5)). A key assumption here, standard in DM (Marantz 1997), is that they contain, at a minimum, a root that is unspecified for features traditionally associated with nouns ( person, number, case, etc.) and a categorydefining node n: [root n]. In addition, I use K as an umbrella for what is realized as case (see, e.g., Caha 2009, Radkevich 2010, Pesetsky 2013, for more articulated representations). Similarly, I collapse the φ-features, and for expository reasons I equate φ with its internal constituents, in particular with number (#). Furthermore, in accordance with Greenberg’s (1963) universal 39, K is assumed to be located higher than #.
Universal 39 (Greenberg 1963:95)
Where morphemes of both number and case are present and both follow or both precede the noun base, the expression of number always comes between the noun base and the expression of case.
This gives the abstract representation for a canonical lexical noun in (5). As alluded to above, the crucial characteristic that distinguishes lexical nouns from pronouns is that the former contain a root and a category-defining n. Specifically, the presence of n results in K’s being insufficiently local to the root to govern its suppletion (see in particular Embick 2010 and Bobaljik 2012 for locality restrictions in DM).
Minimally, the cyclicity hypothesis is assumed, which entails that accessibility to structure is domain-dependent. That is, certain nodes in the structure function as domain delimiters and morphological processes are confined to operate within domains. In syntax, cyclic domains are implemented as phases (Chomsky 2000, 2001), where the head of a phase triggers Spell-Out of its sister, rendering elements in the sister opaque to interactions with elements in higher domains. The analogous notion within morphology is that there are locality domains within a complex X0. A domain-delimiting head induces the morphological analogue of Spell-Out, VI, of its sister. In a complex X0 [[A α] B], if α is a cyclic head, it spells out its sister: A. Assuming that Spell-Out freezes a string, B and A cannot interact across α (Embick 2010, Bobaljik 2012; see Scheer 2010 for an overview). In morphology, domain delimiters are category heads (Embick 2010).
On the assumption that a category head is a cyclic head that spells out its sister, n causes Spell-Out of the root, which here amounts to VI of the root.5 Now, if Spell-Out and accessibility for governing suppletion lined up perfectly, no allomorphy would ever cross a category- defining node, since the root would always be closed off. However, this theory would be too restrictive, incorrectly limiting suppletion to be exclusively sensitive to category-defining nodes, whereas in fact there is plenty of evidence that the root must have access to at least a small amount of structure above the domain-defining head (Embick 2010).
Following Embick (2010), I assume that a morphological dependency may span no more than one cyclic node. In effect, this amounts to the following: in a structure [[A α] B] where α is cyclic, although only A is subject to Spell-Out, both (cyclic) α and B are accessible to condition contextual allomorphy at A ( provided that α is phonologically null, since Embick adds linear adjacency as a requirement on contextual allomorphy). There are at least two ways this condition might be derived. On the one hand, it might simply be stipulated, a kind of morphological subjacency (see Moskal to appear), along the lines of Embick’s (2010) C1LIN (see also section 5). On the other hand, it might be derived from a dynamic approach to domains. Specifically, Bobaljik and Wurmbrand (2013) argue that nodes are designated inherently only as potential phase initiators; whether a node is an actual phase initiator or not depends on whether it is the highest node of an extended projection (Grimshaw 2005; see also Bobaljik and Wurmbrand 2005, Den Dikken 2007, Bošković 2014, Wurmbrand 2014). That is, not until the next node is accessed is it known whether the potential phase initiator is the top projection and, as such, an actual phase initiator. Since access to one node above the category-defining node is required to determine its status, that node itself is a potential context for root allomorphy, and therefore, the accessibility domain is made up of the VI node and one node above.
Returning to the structure of lexical nouns (5), recall that VI proceeds cyclically from the root outward (Bobaljik 2000, Embick 2010). Cyclic n triggers Spell-Out (here, VI) of its complement, the root. For practical application, consider the VI rules for the suppletive forms for ‘child’ in Ket (2), illustrative of the general schema for number suppletion.6
The more specific VI rule is selected, √CHILD ⇔ kΛʔt / _____ PL, and the # value is available to condition root suppletion since at the point where the root undergoes VI, # is sufficiently local to govern root suppletion. That is, when the plural is merged, n (a potential domain) is confirmed as a domain, and VI applies at the root, with n and # being accessible.
Crucially, the root cannot access information about case, however, since at the point where the root is subject to VI, K is located too far away to govern root suppletion. It is important to note that it is cyclic locality that prevents the root from accessing case information. That is, nothing prevents the formulation of a hypothetical VI entry making reference to K, as in (11); rather, a VI rule as in (11) is an illegitimate item, since K is inaccessible for reasons of locality.
(11) √CHILD ⇔ gu: / _____ K
At this point, a note on portmanteaux is in order (e.g., Halle and Marantz 1993, Radkevich 2010). In languages in which # and K are fused into a single morpheme (a portmanteau), we might predict that K should be able to influence root suppletion since the node hosting the #+K complex seems sufficiently local.
Crucially, though, I assume that (12) is not the underlying structure; rather, it is derived from (5). Therefore, although portmanteaux look like they have the structure in (12), at the relevant point of the derivation, they still have the structure in (5). Specifically, if their portmanteau structure is only relevant when (at least one of ) their constituents undergo VI, the portmanteau nature of [#-K] would only be relevant when (at least) # undergoes VI. In this way, portmanteau creation counterfeeds case-driven root suppletion since the former applies too late. Crucially, at the point where the root is subject to VI, it is irrelevant whether the node that is structurally adjacent, #, will later become part of a portmanteau or not; either way, the locality restrictions hold and case-driven suppletion is banned.8
In sum, while number-driven root suppletion is possible, casedriven suppletion is excluded by cyclic locality. Thus, we derive the lack of case-driven root suppletion in lexical nouns.
With regard to pronouns (6), recall that I assume they lack n. Importantly, the absence of n has the result that in pronouns both # and K are potential contexts for suppletion. Concretely, the VI entries for Latvian (singular) 1st person (4) are as follows:
 ⇔ man / _____ K
 ⇔ es
Crucially, VI entries for pronouns that make reference to K are legitimate items since K is accessible. That is, given the lack of a category-defining node, no domain is created low in the structure: in pronouns, suppletion in the context of number as well as case is possible.
4 More or Less Structure
4.1 Numberless Nouns
An interesting prediction from the proposal made here is that in nouns that lack a # node, case-driven root suppletion should become possible.9 In Archi (Nakh-Daghestanian), the form for ‘father’ displays suppletion for ergative case (Archi Dictionary).
Intriguingly, though, this form is a singulare tantum and as such does not have a corresponding plural. I propose that Archi’s ‘father’ is defective in that it lacks a # projection and its inherent singular value is located on n (see Kramer 2012 and Smith 2014 for proposals that inherent number may be on n). Given that # is missing in Archi’s ‘father’, the (ergative) K node is accessible as a context for root suppletion since K becomes the node above category-defining n. The VI rules for (15) are given in (17).
√FATHER ⇔ úmmu / _____ K
√FATHER ⇔ ábt:u
Crucially, the VI entries in (17) are interpretable since K is accessible for this item.
In addition to Archi’s ‘father’, the form for ‘child’ in Archi and ‘water’ and ‘son’ in Lezgian (Nakh-Daghestanian) also supplete for case.10 In Moskal to appear, I argue that these three nouns lack a # node in the relevant context, thus allowing for (limited) case-driven root suppletion. For instance, consider Lezgian ‘water’ (Haspelmath 1993, pers. comm.).
First, note that the plural morpheme (-ar) blocks the suppletive root from surfacing ( jat-ar-i ‘water-PL-OBL’ rather than *c-ar-i). Indeed, we expect the regular root to surface in the oblique plural, since the plural morpheme intervenes between the root and K, blocking root suppletion (see also section 4.2): [[[√WATER n] PL] OBL].
Second, the suffix on the singular oblique form is a single vowel -i, which I analyze in Moskal to appear as a ‘‘pure’’ oblique suffix. Other nouns have an additional affix between the root and the oblique marker, in the singular, which I take to be an exponent of singular number: fíl-d-i ‘elephant-SG-OBL’. Therefore, I suggest that in the item for ‘water’, the singular is pruned in the context of oblique case, resulting in a structure analogous to that of Archi’s ‘father’: [[√WATER n] OBL].11
A second prediction that follows from the locality restrictions identified here is that when a node X intervenes between n and #, casedriven root suppletion should be blocked, since only n and X would be accessible to potentially govern suppletion.
In Slavic languages, the diminutive is closer to the root than #: [[[root n] DIM] #]. Strikingly, as the following Serbo-Croatian case shows, this configuration indeed prevents case-driven root suppletion.
Rather than having a suppletive form ljudići, Serbo-Croatian forms the diminutive plural with a periphrastic construction, mali ljudi ‘small people’ (čovečići is marginally accepted). Similarly, in Polish the suppleting singular-plural pair for ‘man’, człowiek – ludzie, fails to supplete in the diminutive: *ludziki (if anything, człowiecz-k-i is accepted). Note that Polish does have a form ludz-ik-i, but this refers to figurines and has a corresponding singular ludz-ik ‘figurine-DIM’. In Russian, the form čelovečki is attested in the Russian National Corpus, whereas ljudčiki is not.
In sum, when an element intervenes between n and #, numberdriven root suppletion is blocked, exactly as predicted by the formulation of locality assumed here.12
It is worth noting that Arregi and Nevins (to appear) account for the lack of suppletion in one form of a disuppletive pair, such as worse/badder, by drawing on a parallel notion of blocking where an additional functional head intervenes in the structure. For example, suppletion is blocked in the comparative form of evaluative bad— namely, badder—owing to the presence of an evaluative element, adding further support to locality as defined here.
5 Final Remarks
In this squib, I addressed the peculiar asymmetry between lexical nouns and pronouns with regard to case-driven suppletion. I showed that a minimal approach to locality, which crucially draws on syntactic hierarchical structure as the input to morphology, suffices to explain why case-driven suppletion fails to be observed in canonical lexical nouns, while it is common in pronouns. Specifically, given a dynamic approach to locality, the presence of a category-defining node n induces a domain low in the structure, preventing K from being sufficiently local to govern root suppletion. Furthermore, nodes that are accessible for governing root suppletion are the category-defining ( potentially domain-inducing) node as well as the node above it.
One last issue that warrants attention is how the current proposal compares with Embick’s (2010) approach to locality. Drawing on (a version of ) the Phase Impenetrability Condition (PIC), Embick proposes that a phasal head β induces Spell-Out not of its own complement but of the complement of phasal head α lower in the structure than β (where β itself is not accessible as an allomorphy restrictor). Applied to nouns, the ban on case-driven root suppletion would require that, in addition to n, some other cyclic node would need to be present in the structure, crucially located above n. Indeed, on an approach where K was cyclic (e.g., by virtue of being the highest node), K would induce Spell-Out of the root but would not be accessible. However, numberless nouns provide the crucial argument against cyclic K: the lack of (noncyclic) # should have no effect on locality and K should still be inaccessible, contrary to fact. Alternatively, as an anonymous reviewer suggests, D would be a plausible candidate; however, besides the fact that cyclic D faces the same problem as cyclic K, it is controversial whether D is universally present (e.g., Bošković 2008). In contrast to Embick’s approach, the approach taken here naturally accommodates the possibility of numberless nouns suppleting for case, and relies on fewer stipulations about which nodes are domain delimiters, (currently) only committing to category-defining nodes.
In sum, this study bears on the formalization of locality domains as employed in DM. The hypothesis advocated here relies on (morpho-) syntactic structure playing a crucial role in the discrepant behavior, which raises the question whether these observations can be captured in frameworks that give no role to hierarchical structure in the morphology.
Appendix A: Languages That Display Root Suppletion in the Context of Number
!Xóõ (Khoisan), Afrikaans (Indo-European), Arapesh (Torricelli), Archi (North-East Caucasian), Dinka (Nilo-Saharan), Eastern Pomo (Pomoan), Gaelic (Indo-European), Hebrew (Afro-Asiatic), Hopi (Uto-Aztecan), Hua (Trans-New Guinea), Ket (Yeniseian), Khakas (Altaic), Komi (Uralic), Lango (Nilo-Saharan), Lavukaleve (Central Solomons), Lezgian (North-East Caucasian), Russian (Indo-European), Tariana (Arawak), Tiwi (isolate), Turkana (Nilo-Saharan), Yimas (Sepik-Ramu), Zulu (Niger-Congo)
Appendix B: Languages Included in the Study
!Xóõ (Khoisan), Afrikaans (Indo-European), Akwesasne Mohawk (Iroquoian), Arapesh (Torricelli), Archi (North-East Caucasian), Basque (isolate), Bilua (Central Solomons), Boumaa Fijian (Austronesian), Burushaski (isolate), Cahuilla (Uto-Aztecan), Carib (Cariban), Cavinen ˜a (Tacanan), Crow (Siouan), Dagaare (Niger-Congo), Dinka (Nilo- Saharan), Dolakha Newar (Sino-Tibetan), Dumi (Sino-Tibetan), Dzongkha (Sino-Tibetan), Eastern Pomo (Pomoan), Evenki (Altaic), Finnish (Uralic), Gaelic (Indo-European), Georgian (Kartvelian), Hebrew (Afro-Asiatic), Hopi (Uto-Aztecan), Hua (Trans-New Guinea), Hungarian (Uralic), I’saka (Skou), Itelmen (Chukotko-Kamchatkan), Itzaj Maya (Mayan), Jacaltec (Mayan), Japanese (Japonic), Jarawara (Arauan), Jingulu (Australian), Kannada (Dravidian), Kashmiri (Indo- European), Kayardild (Australian), Ket (Yeniseian), Khakas (Altaic), Kham (Sino-Tibetan), Kiowa (Kiowa-Tanoan), Klon (Trans-New Guinea), Koasati (Muskogean), Komi (Uralic), Koromfe (Niger- Congo), Kwaza (isolate), Ladakhi (Sino-Tibetan), Lango (Nilo-Saharan), Latvian (Indo-European), Lavukaleve (Central Solomons), Lezgian (North-East Caucasian), Malayalam (Dravidian), Manam (Austronesian), Mangarayi (Australian), Maori (Austronesian), Mapuche (Mapudungu), Mayali (Australian), Maybrat (Maybrat), Mina (Indo-European), Modern Khwe (Khoisan), Mosetén (Mosetenan), Nahuatl (Uto-Aztecan), Nishnaabemwin (Algonquian), Paraguayan Guaraní (Tupian), Puyuma (Austronesian), Rabha (Sino-Tibetan), Russian (Indo-European), Semelai (Mon-Khmer), Sinaugoro (Austronesian), Tamil (Dravidian), Tariana (Arawak), Thai (Tai-Kadai), Tiwi (isolate), Turkana (Nilo-Saharan), Turkish (Altaic), Vietnamese (Austro-Asiatic), Yanyuwa (Australian), Yimas (Sepik-Ramu), Zulu (Niger-Congo)
Many thanks to Jonathan Bobaljik, Andrea Calabrese, and Peter Smith for invaluable discussion of the ideas presented here. I would also like to thank audiences at BLS 39, GLOW 36, PhonoLAM, SIRG (McGill/UQAM), and the Allomorphy: Logic and Limitations workshop, and the anonymous LI reviewers for all their comments.
1 The important question of exactly what counts as a suppletive root cannot be resolved here. Rather, the criterion for noun suppletion here is singularplural pairs identified as suppletive in prior literature, where these are strongly suppletive—that is, not plausibly related by ( possibly idiosyncratic) phonological (readjustment) rules.
2 See appendix A for a (nonexhaustive) list of languages with numberdriven noun suppletion.
3 As pointed out by two anonymous reviewers, nouns and pronouns also differ in terms of frequency as well as their diachronic history. However, while diachrony and frequency effects relate to typological generalizations, locality effects as proposed here are assumed to be universal (see Corbett 2007) and as such allow no exceptions; see Kiparsky 2008 for discussion. Also recall that one of the asymmetries identified here involves nouns alone: while noun suppletion is well-attested, it is only in the context of number, not of case.
4 Two notes are in order here. First, in (5)–(6) I use standard adjunction structures, with X0 [X0 Y0]. All tree diagrams here and below will represent internally complex X0s, however those are derived (see Bobaljik 2012). The labeling of the nonterminal (branching) nodes of the structures will not play a role in the analysis.
Second, De Belder and Van Craenenbroeck (2011) suggest that pronominal exponents can also be inserted in root position, rather than always occurring as true pronouns, in order to account for forms such as ge-ik ‘egocentricity’ (Dutch) and duzen ‘to address with the familiar 2nd person singular form’ (German). Indeed, following this line of thought, these particular forms should fail to supplete for case, which is correct (*ge-mij, *dirzen).
6 Here I put aside the question of when the plural morpheme is the regular plural exponent or a zero, an issue that arises in English past tense (run–ran vs. tell–tol-d) and comparatives (bett-er vs. worse) as well.
7 Prima facie, Slovenian offers a counterexample: in the dual, člóvek ‘person’ suppletes for genitive and locative; however, genitive dual and locative dual are syncretic with genitive plural and locative plural, respectively (Corbett 2009). Thus, we observe plural-driven rather than case-driven suppletion.
9 It is the presence or absence of # rather than the overt realization of number morphology that determines K’s accessibility for governing root suppletion (see Moskal 2013 for discussion).
11 I assume that there is a preference to keep nodes intact. Indeed, work on (adjectival) suppletion (Bobaljik 2012) seems to show that the lack of an exponent does not change structural relations; that is, if pruning of (null) Cmpr were freely available, then Sprl would be expected to trigger root suppletion, which is unattested. Thus, pruning seems to be a highly marked configuration; presumably, missing features are derivationally costly, but I leave an investigation into the nature of pruning to future research.
12 As an anonymous reviewer points out, the current proposal makes predictions about more fine-grained morphological subanalysis; I leave this to future research.