Much recent literature has focused on whether the verbal agreement morphology cross-referencing objects is true φ-agreement or clitic doubling. I address this question on the basis of comparative data from related Inuit languages, Inuktitut and Kalaallisut (West Greenlandic), and argue that both possibilities are attested in Inuit. Evidence for this claim comes from diverging syntactic and semantic properties of the object DPs encoded by this cross-referencing morphology. I demonstrate that object DPs in Inuktitut display various properties mirroring the behavior of clitic-doubled objects crosslinguistically, while their counterparts in Kalaallisut display none of these properties, indicating genuine φ-agreement rather than clitic doubling. Crucially, this distinction cannot be detected morphologically, as the relevant cross-referencing morphemes are uniform across Inuit. Therefore, this article cautions against the reliability of canonical morphological diagnostics for (agreement) affixes vs. clitics.
Verbal agreement morphology is commonly analyzed as the morphological reflex of φ-feature valuation of a probing head H0 by a φ-bearing goal, the result of Agree (e.g., Chomsky 2000, 2001). However, much recent literature has argued for a contrast between the agreement morphemes cross-referencing subjects and those cross-referencing objects: while subject agreement is often considered to be genuine φ-agreement, many cases of apparent object agreement have been reanalyzed as pronominal clitic doubling (e.g., Woolford 2008, Preminger 2009, Nevins 2011, Kramer 2014, Anagnostopoulou 2016). Unlike true agreement, clitic doubling involves a pronominal D0 element cooccurring and coreferring with a DP associate. The clitic and its associate are moreover often claimed to be related via a movement chain (Torrego 1988, Uriagereka 1995, Anagnostopoulou 2003, Arregi and Nevins 2012, Harizanov 2014, Baker and Kramer 2016, 2018). The structural difference between φ-agreement and clitic doubling is schematized in (1).
Concomitantly, there has been much discussion on how to identify whether a given φ-bearing morpheme results from true agreement or clitic doubling. Some of the authors cited above have suggested morphosyntactic diagnostics for distinguishing the structures in (1) (e.g., Kramer 2014, Baker and Kramer 2016, 2018). For example, if the morpheme in question is a doubled clitic, then it should behave like a pronoun or determiner, and behave as though it is linked to its associate by movement. In contrast, if the morpheme is genuine φ-agreement, then it is not expected to have such properties. At the same time, it has also long been assumed that true φ-agreement displays morphological properties typical of “affixes,” while doubled clitics crucially do not (Zwicky and Pullum 1983, Nevins 2011, Compton 2016). However, as these latter criteria are often heuristic in nature, they are ultimately only contentful if the morphological properties in question correlate with other independent factors that differentiate φ-agreement and pronominal clitics.
This article presents a case study showing that the underlying status of this object-referencing morphology should be determined without appealing to its surface appearance. The argument comes from variation within the Inuit (Eskimo-Aleut) dialect continuum, in which object-referencing morphology targets ABS (absolutive) objects. I identify a constellation of grammatical differences between two Inuit varieties—Kalaallisut (West Greenlandic) and Inuktitut—and argue that this follows from a dialectal split between object φ-agreement (in Kalaallisut) and pronominal clitic doubling (in Inuktitut). In Kalaallisut, ABS object DPs raise to a structurally high position (Spec,AgrOP, to be motivated later) and are cross-referenced by genuine φ-agreement, (2a) (Bittner 1994, Bittner and Hale 1996b, Woolford 2017). In contrast, on the basis of novel fieldwork, I propose that the raising element in Inuktitut is not a full DP, but a φ-bearing D0doubling anABSobject DP, (2b).
Crucially, the object-referencing forms in Kalaallisut and Inuktitut are morphologically uniform, and display properties typically associated with affixes and not morphophonological clitics. Appealing to morphological appearance alone would thus obscure the structural difference argued for here.
Evidence instead comes from an array of ABSobject asymmetries in Inuktitut, in that ABS objects pattern distinctly from all other arguments, including ABS subjects. As I will show, the exact constellation of properties seen in Inuktitut is highly reminiscent of clitic-doubled objects crosslinguistically. Thus, whereas the verbal morphology encoding ERG (ergative) and ABS subjects in Inuktitut is true φ-agreement, its ABS object-referencing counterpart is best analyzed as a pronominal D0. In contrast, these asymmetries are wholly absent in Kalaallisut, suggesting that Kalaallisut lacks object clitic doubling—thus, its object-referencing morphemes are genuine φ-agreement, on par with its subject-referencing morphemes. Finally, as independent support for this proposal, I show that this split is not arbitrary, but may be subsumed within broader variation in the degree of object shift across the Eskimo-Aleut language family, building on an insight of Woolford’s (2017).
The article is organized as follows. In section 2, I outline previous literature on the distinction between object φ-agreement and clitic doubling. In section 3, I provide an overview of Inuit morphosyntax, paying particular attention to the properties of the object-referencing morphemes under investigation. In sections 4–5, I argue that these morphemes in Inuktitut are the product of pronominal clitic doubling, and, concomitantly, that the analogous morphemes in Kalaallisut are genuine φ-agreement. I also develop an analysis of object clitic doubling that accounts for the morphological idiosyncrasies found in Inuktitut. Finally, in section 6 I show how this analysis sheds light on variation in object shift across Eskimo-Aleut.
2 Object Agreement vs. Clitic Doubling
Clitic doubling is the cooccurrence of a reduced pronominal element (a clitic) with a full DP. Unlike in clitic left- or right-dislocation, the full DP is in its base position, not dislocated, and the doubled clitic is generally optional. This article focuses on the clitic doubling of direct objects.1 A canonical example of clitic doubling is given in (3), from Romanian.
(Il) văd pe Ion.
him.CL see.I PE John
‘I saw John.’
It has been claimed that many (or perhaps even all) cases of what has been taken to be object φ-agreement should actually be analyzed as clitic doubling (Woolford 2008, Nevins 2011). According to these proposals, these morphemes have the same underlying structure as the Romanian clitic in (3), despite surface appearances. This idea stems from the observation that, even in languages in which the object-referencing morphology does not look clitic in nature, it nonetheless displays the syntactic and semantic properties otherwise characteristic of clitic doubling (e.g., Riedel 2009, Kramer 2014, Anagnostopoulou 2016, Baker and Kramer 2016). On the other hand, it has been argued that there does exist object φ-agreement that is distinct from clitic doubling (e.g., Oxford 2014). The status of object-referencing morphology has also been debated in the context of the Inuit languages (Compton 2016, Johns 2017, Johns and Kučerová 2017).
Traditionally, diagnostics to distinguish genuine φ-agreement from pronominal clitic doubling have focused on morphological tendencies that distinguish affixes and morphophonological clitics, rather than sentence-level properties (Zwicky and Pullum 1983, Woolford 2008, Nevins 2011). For example, Zwicky and Pullum (1983) argue that agreement morphemes often display allomorphy and morphological irregularities, while clitics are expected to be invariant; moreover, while clitics are able to attach to stems that contain affixes, affixes cannot attach outside of clitics. However, as noted above, these distinctions are not universal and do not obviously follow from any theoretically grounded differences between the two. Thus, it is not clear that there is a reliable link between affixes and agreement, and between morphophonological clitics and pronominal clitics.2
For this reason, Nevins (2011) offers an alternative diagnostic based on contextual morphological variance, proposing that, since agreement expones φ-features on some functional head (e.g., T0), it may interact with other features on that head. Conversely, he suggests that clitics—D0s that adjoin to their host (again, T0)—are expected to be invariant.3 This contrast is borne out in Spanish, in which subject φ-agreement is tense-variant, but object pronominal clitics are not.
However, this diagnostic faces the same issues that Nevins identifies for Zwicky and Pullum’s (1983) proposal. It is not obvious how to rule out this type of variance between a pronominal clitic and its host within a postlexicalist framework such as Distributed Morphology,4 as these elements are structurally immediately adjacent (see (1b)). Furthermore, allomorphy between a verbal head and an adjacent pronoun is in fact attested in natural language. For instance, as summarized in Bennett, Elfner, and McCloskey 2019, different varieties of Irish display alternations of this very sort. In the Donegal and Mayo dialects, the forms of various verbal morphemes are conditioned by the presence of a subject pronoun (analyzed by Bennett, Elfner, and McCloskey as a bare D0). The reverse holds in the coastal Munster dialects, in which the form of the subject pronoun itself is conditioned by verbal inflection. These effects—especially the latter—contradict both Zwicky and Pullum’s (1983) and Nevins’s (2011) proposals.
Thus, I contend that it is more fruitful to frame diagnostics for clitic doubling around the structural and derivational relationship between the clitic and its associate. For example, if clitic doubling involves a syntactic dependency between a D0 and a coindexed DP, then the pronominal status of the clitic should have consequences for the distribution and interpretation of the DP associate (see Baker and Kramer 2016).5 A novel empirical argument for this idea will come from variation within Inuit.
3 Properties of the Inuit Case and Agreement System
The Inuit languages are a continuum of dialects from the Eskimo-Aleut language family, spoken across the North American Arctic and Greenland. The data in this article represent Kalaallisut (Greenland), based on previous work by Fortescue (1984), Bittner (1994), Berge (2011), and others, and Inuktitut (Nunavut), based on work by Beach (2011) and on my own fieldwork.6 The Inuit languages are polysynthetic, and the order of morphemes within a word generally adheres to the Mirror Principle. A word begins with a leftmost root, followed by a series of suffixes corresponding to successively higher heads along the functional spine. Because agreement morphemes occupy the right edge of the verb complex and are morphologically conditioned by mood/clause type, they are taken to be located in the extended CP-domain (Johns 2007, Compton 2016).7
‘You did not want to eat it.’
‘Did you open it early?’
The agreement forms in (5) target both the subject and the object, which display an ERG-ABS case patterning, shown in (6a). In intransitive contexts, only the ABS subject is cross-referenced, (6b). Finally, the transitive ergative construction alternates with an antipassive construction, in which the transitive subject is ABS and the object bears MOD (“modalis”) case, (6c); here, the MOD object does not appear with agreement morphology, which can only encode ERG and ABS arguments (Murasugi 1994, Bobaljik 2008).8
The agreement forms often (though not always) surface as portmanteaux; in such cases, they do not separately encode the φ-features of the subject and object, as indicated in (5a) and (6a). Comparing the declarative and interrogative forms in (5), we also see that mood/clause type may idiosyncratically trigger suppletive allomorphy on the adjacent agreement morpheme(s) (e.g., Fortescue 1984, Dorais 1986).9 The nearly identical forms from Kalaallisut in (7) show that these properties hold across Inuit.10
(7) Kalaallisut forms
Compton (2016) argues that the nonpredictability of these forms is expected of agreement markers, not pronominal clitics. Moreover, he observes that the mood-sensitive allomorphy passes Nevins’s (2011) tense variance diagnostic—instantiated in Inuit as mood variance. As this variance is able to affect the collective realization of subject and object features in the portmanteau forms, Compton concludes that the object-referencing portions of these agreement complexes cannot be clitic in nature.11
However, a close comparison between Kalaallisut and Inuktitut reveals several previously unnoticed differences concerning the distribution and interpretation of ABS objects—the arguments cross-referenced by object-referencing morphology. I argue that the exact constellations of properties reveal the object φ-agreement vs. clitic-doubling distinction argued for here. Crucially, this cannot be determined by examining the morphemes themselves—which, as shown above, are uniform in both varieties.
4 The Interpretation of ABS Objects across Inuit
It is often noted that ergative and antipassive constructions across Inuit display distinct semantic properties. While the exact effect is somewhat difficult to pinpoint, I will follow Bittner (1994) and Bittner and Hale (1996a,b) in characterizing it in terms of scope, that is, correlating it with syntactic height.12 Regardless, it is almost universally assumed that the locus of the distinction differentiates ABS arguments (both subjects and objects) from MOD objects. However, I demonstrate below that these properties are true for Kalaallisut—but do not extend to Inuit as a whole.
4.1 ABS Arguments in Kalaallisut
In Kalaallisut, the contrast described above can be illustrated by comparing the interpretation of ABS arguments with that of other elements in the sentence. In (8), we see that ABS subjects and objects obligatorily scope above negation, while MOD objects obligatorily scope below negation. According to Bittner (1994:138), this also holds for other sentential operators (e.g., -tariaqar ‘must’, -juannar ‘always’).
Various authors explain this difference as a consequence of movement of the ABS argument to a structurally high position (see (9a–b)), taking scope above sentential negation (e.g., Bittner 1994, Bittner and Hale 1996a,b, Manga 1996, Woolford 2017; cf. Diesing 1992).13 In contrast, antipassive (MOD) objects remain in situ within the VP domain (see (9c)). Under this view, the interpretive contrast between ABS and MOD objects is fed by the syntax, as the scope of a given argument is determined by its structural height. As observed by Woolford (2017), this is reminiscent of object shift in Icelandic, which is optional for DPs and has similar semantic consequences; we will return to this parallel in section 6. Finally, the idea that ABS subjects and objects raise to a uniformly high position, above the locus of non-ABS arguments, fits with the syntactically ergative profile of the Inuit languages.
Another relevant property concerns the distribution of negative polarity items (NPIs). Under negation, a negative indefinite can be created by attaching the disjunctive clitic =luunniit ‘or’ to a quantifier or wh-indeterminate (Bittner 1994, Hallman 2008). Bittner (1994) additionally shows that the licensing of this NPI is sensitive to c-command. The examples in (10) show that, as long as the c-command requirement is met, the NPI may surface in any argument position, including ABS object position (10b). Since ABS arguments in Kalaallisut otherwise exhibit wide scope, Bittner (1994) proposes that, while ABS NPIs undergo Ā-movement to the clausal left periphery, they reconstruct at LF in order to be licensed by negation.
I now show that these properties of Kalaallisut cannot be replicated in Inuktitut.
4.2 ABS Object Asymmetries in Inuktitut
On the surface, the Inuktitut examples in (11) are morphosyntactically analogous to the Kalaallisut examples in (8). However, consider the available interpretations relative to the quantificational adverb qautamaat ‘every day’. The ABS subject and MOD object may both be interpreted as taking apparent wide scope or narrow scope relative to each other, (11a–b). In contrast, the ABS object in (11c) appears to permit only the wide scope interpretation. Moreover, though he does not provide the relevant data, Beach (2011:62–63) states that the same pattern can be seen with sentential negation. The fact that ABS subjects and ABS objects pattern differently provides a first clue that the analysis of Kalaallisut described in section 4.1 cannot be extended to Inuktitut.14
Evidence from NPI licensing in Inuktitut corroborates this point, and moreover reveals that the wide-scope-like effect seen in (11c) does not arise from structural height—again, in contrast to Kalaallisut. In Kalaallisut, we saw that NPIs may surface in all argument positions, including ABS object position; recall, moreover, Bittner’s (1994) analysis based on syntactic movement and LF reconstruction. In contrast, (12) shows that, in Inuktitut, the same NPI may appear in any position exceptABS object position.15
Given Bittner’s (1994) analysis of Kalaallisut, we might initially want to analyze the ill-formedness of (12c) as an exceptional inability for structurally high ABS objects to reconstruct. However, this cannot be the case: NPIs trapped in syntactic islands (ruling out covert movement past matrix negation) still may not surface in ABS object position.16
In summary, we have seen that, whereas ABS subjects and objects pattern together in Kalaallisut, Inuktitut displays ABS object asymmetries. Moreover, whereas ABS arguments in Kalaallisut can be uniformly analyzed as raising to a structurally high position, the exact nature of the asymmetries in Inuktitut indicates that the exceptional semantic behavior of ABS objects requires an alternative explanation. I will argue below that the key difference lies in the object φ-agreement vs. clitic-doubling distinction, which shapes the degree of ABS object movement across Inuit.
4.3 Object Clitic Doubling in Inuktitut
The fact that ABS objects seem to be exceptional in Inuktitut requires an analysis that isolates this particular combination of case and argument position from the others in the language (e.g., ABS subject, MOD object). Now observe that ABS objects happen to be the only nominals cross-referenced by object agreement morphology. I propose that this is the relevant factor, and the semantic asymmetry arises because object-referencing morphology in Inuktitut is underlyingly pronominal clitic doubling. In contrast, because Kalaallisut lacks these semantic asymmetries, I conclude that the surface-equivalent morphemes in Kalaallisut expone genuine object φ-agreement.
The kinds of effects observed in Inuktitut are robustly attested crosslinguistically: clitic-doubling languages generally forbid clitic doubling of nonreferential, nonspecific objects, including negative indefinites, but require clitic doubling of referential (i.e., D-linked) or specific objects (e.g., Suñer 1988, Dobrovie-Sorin 1990, Franks and Rudin 2005, Kramer 2014, Baker and Kramer 2016, 2018). Examples from Romanian are provided in (14).17
Although there is much debate over how to characterize and derive the semantic contribution of clitic doubling, this may be due to crosslinguistic variation in the exact effect that arises (see section 5.1). Regardless, the directionality of the effect appears to be universal: clitic-doubled nominals tend to have special interpretations reminiscent of those associated with pronouns or definite determiners, while nominals that are not doubled do not.
Indeed, this pattern is also observable in Inuktitut. We saw in section 4.2 that Inuktitut ABS objects appear to take wide scope and are incompatible with NPI licensing. Below, I additionally show that wh-phrases in ABS object position in Inuktitut are obligatorily D-linked, thus following the same general pattern.18 Whereas simplex wh-phrases (e.g., kisu ‘what’) need not be interpreted as D-linked in ABS subject and MOD object position, (15a–b), the same phrases are interpreted as D-linked in ABS object position, (15c).
Crucially, this effect is obligatory, as illustrated with aggressively non-D-linked arguments19 in Inuktitut (e.g., ‘wh-the-hell’ or ‘wh-on earth’), expressed with the vagueness enclitic =kiaq. As predicted, they are ruled out in ABS object position.
In summary, I have shown that Inuktitut displays a semantic asymmetry that diverges from Kalaallisut. Whereas ABS subjects and objects pattern together in Kalaallisut, in Inuktitut ABS objects contrast with all other arguments, including ABS subjects. On the basis of crosslinguistic parallels with better-studied languages, I proposed that Inuktitut ABS objects are clitic-doubled. Below, I develop an explicit derivation of clitic doubling that accounts for both the Inuktitut facts shown above and their morphological similarities with Kalaallisut.
5 The Morphosyntax of Object Clitic Doubling
5.1 Derivation of Clitic Doubling
Building on work by Baker and Kramer (2016, 2018), I propose that object clitic doubling in Inuktitut involves three steps. First, AgrO0 enters into an Agree relation with the ABS object DP, triggering phrasal movement to Spec,AgrOP in the clausal left periphery, (17a).20 This first step takes place in both Inuktitut and Kalaallisut. However, in Inuktitut the DP undergoes an additional syntactic operation—termed Reduce by Baker and Kramer—which converts the DP into a pronominal D0, (17b). M-Merger then applies postsyntactically, rebracketing the specifier-head configuration in (17b) into a complex head, (17c) (Matushansky 2006).21
It is crucial that the operation that converts the raised DP to a bare D0 takes place in the syntactic component. This contrasts with Harizanov’s (2014) proposed DP→D0 process of clitic doubling, which is postsyntactic. These approaches can be differentiated in their predictions for the interpretation of object clitic doubling at LF. Under the present analysis, because the clitic is created syntactically, this element is also interpreted as a D0 at LF. In contrast, if the clitic is generated postsyntactically, then the raised element should be interpreted as a full DP at LF. Indeed, one motivation for Harizanov’s (2014) account is to draw syntactic and semantic parallels between object clitic doubling and object shift—both understood as DP-movement under his analysis.
However, this approach is not viable for Inuktitut, given the contrast with Kalaallisut identified above. If the uniform interpretation of ABS subjects and objects in Kalaallisut is derived from syntactic movement of a DP (i.e., object shift), then the behavior of ABS objects in Inuktitut cannot also be derived by this process.
A syntactic operation like Reduce instead ensures that the interpretation of clitic-doubled ABS objects is distinct from that of genuinely structurally high elements, like ABS subjects. Building on work by Suñer (1988), I propose that the special interpretation of clitic-doubled objects arises from a Matching Principle imposed between the D0 and its DP associate; the two must match in both φ-features and semantic features because they are coindexed members of a chain. In Inuktitut, pronominal D0s are standardly interpreted as referential, (18). Therefore, quantificational DPs must be interpreted as D-linked when doubled by a pronominal D0.
Context: Jaani took a pencil from the pencil case . . .
amma tuni-janga Miali-mut.
and give-3SG.S/3SG.O Miali-ALLAT
‘ . . . and gave it to Miali.’
Context: David bought a new shirt . . .
#amma=lu Kiuru-p niuvi-lau-mmi-janga.
and=CONJ Kiuru-ERG buy-PST-also-3SG.S/3SG.O
‘ . . . and Kiuru bought it too.’ (Unavailable: ‘ . . . and Kiuru bought one too.’)
As independent evidence, Runić (2014) has observed that languages whose pronominal clitics are semantically flexible, such as Serbian (see (19a)), do not impose special interpretive effects on their doubled associates. Under the present analysis, this is because the Matching Principle can be more easily satisfied. In nonstandard Serbian varieties that permit clitic doubling, clitic-doubled objects can be understood as nonspecific, (19b).
Nikola je vidio film, a vidio ga je i Danilo.
Nikola AUX.3SG saw film and saw it.CL.ACCAUX.3SG and Danilo
‘Nikola saw a movie and Danilo saw it/one too.’
Context: There is a considerable number of old and sick people in the village.
However, there is no doctor in the village.
Opština (ga) novog lekara traži.
municipality him.CL.ACC new doctor look.for.3SG
‘The municipality is looking for a new doctor.’
We are now in a position to derive the morphologically idiosyncratic properties of Inuktitut clitic doubling, as well as their morphological similarities with Kalaallisut. As illustrated below, the mood-agreement complex in Inuit can be schematized as comprising a series of contiguous heads, Mood0, AgrS0, and AgrO0. In Kalaallisut, (20a), both AgrS0 and AgrO0 bear φ-probes, which are valued by the ERG subject and ABS object, respectively. In Inuktitut, (20b), the crucial difference is that the ABS object’s φ-features are exponed by D0, rather than by AgrO0.22
Because D0 forms a complex head with AgrO0 in (20b), standard locality conditions on subsequent morphological operations—such as those responsible for portmanteaux—are met. As one illustration, consider the spanning approach to portmanteaux proposed by Svenonius (2012) and Merchant (2015). Because the relevant heads are contiguous and occur within a single extended projection (here, the extended CP), they may be exponed by a single morph. In Kalaallisut, the relevant span contains the boldfaced heads in (20a); in Inuktitut, the span also includes the pronominal D0.
In sum, I have shown that pronominal clitic-doubling structures are compatible with portmanteaux and other morphological effects, contrary to proposals by both Zwicky and Pullum (1983) and Nevins (2011). Because the pronominal D0 is structurally adjacent to its host (which, in turn, may be adjacent to a subsequent head, and so on), there is no theoretical reason why these heads cannot interact morphologically.
5.2 Pronominal Clitics without Doubling
Whereas the discussion in section 4 uncovered an asymmetry in Inuktitut concerning the interpretation of ABS objects, I now provide additional evidence for Inuktitut clitic doubling based on distributional restrictions on independent pronouns in ABS object position. This, in turn, will help support the idea that clitic doubling is derived by syntactic movement, by delineating the conditions governing copy spell-out in clitic-doubling constructions.
I start by establishing that independent pronouns in Inuit are bare D0s, not phrasal DPs (e.g., Postal 1966, Abney 1987, Elbourne 2005). This treatment is evidenced by the observation that they may function as determiners in so-called adnominal pronoun constructions (APCs)23 such as we linguists. Although APCs have not been studied in detail in Inuit, Fortescue’s (1984) grammar offers (21a), from Kalaallisut.24 In contrast, pronouns in APCs in Inuktitut surface as φ-bearing bound morphemes on the NPs that they modify, (21b); this option seems to be unattested in Kalaallisut.
I take this contrast to indicate that M-Merger of D0 with a structurally adjacent head is a generalized rule in Inuktitut (though not in Kalaallisut), regardless of the exact syntactic environment. In the clausal left periphery, D0 undergoes M-Merger with AgrO0, as shown in section 5.1; in the nominal domain, we see this operation apply in APCs, (21b).
Having established the structure of independent pronouns in Inuktitut, let us now turn to how they interact with clitic doubling. The Inuit languages are generally pro-drop, with the features of the unpronounced pronoun being recoverable from the verbal agreement morphology. However, although Inuktitut allows ERG and ABS subject pronouns to cooccur with subject φ-agreement, (22a–b), ABS object pronouns are forbidden from cooccurring with object clitics. The examples in (22c) are therefore grammatical only if the overt pronoun is suppressed. Though there was some interspeaker variation in the acceptability of (22a–b) (with many speakers finding these examples slightly redundant, though otherwise fine), all speakers consulted judged examples like (22c) as completely ungrammatical. Furthermore, the consultant’s comment in (22c) clearly suggests that the ungrammaticality is due to the cooccurrence of the ABS object pronoun and the object-referencing morpheme.25
Since Kalaallisut is hypothesized to lack pronominal clitics altogether, we predict that independent pronouns may cooccur with verbal agreement in all positions, including ABS object position. Naturally occurring data from Berge 1997 show that this is borne out.
I propose that the asymmetry instantiated in (22) follows from the idea that the pronominal clitic (a D0) and its associate (also a D0 in this context) are identical members of a movement chain, established in section 5.1. Following Landau (2006), the pronunciation of movement chains is constrained by an economy condition that normally triggers deletion of all but one copy. Moreover, the choice of which copy to pronounce is subject to various considerations applying at PF. Together, these derive the cooccurrence restriction in Inuktitut: the higher copy is obligatorily pronounced because it undergoes M-Merger with Agro0; deleting it would violate the Stray Affix Filter (Lasnik 1995). The lower copy is instead deleted to satisfy the aforementioned economy condition. In contrast, clitic doubling permits the D0 and DP to cooccur because they are syntactically distinct.
Of course, object clitics often do cooccur with pronouns crosslinguistically. However, the pronouns in these cases are often emphasized or are otherwise understood as information-structurally salient. This is also true for Inuktitut, in which contrastively focusing an ABS object pronoun (boxed) obviates the cooccurrence restriction.
According to Landau (2006), this is because the economy condition mentioned above may be overridden by other conditions. In (24), this may be attributed to intonational requirements imposed on focus-marked elements. In (25), we further see that pronouncing an ABS object pronoun in its base position is also possible, if it itself undergoes M-Merger with the head of an adjacent element—for example, an adjective, (25a), or a nominal complement in an APC as previously discussed, (25b).
For concreteness, these patterns are schematized in (26).
Again, none of the above discussion is relevant to Kalaallisut, in which bare pronouns may surface in all environments and in pragmatically neutral contexts, including ABS object position. As I suggested, this contrast is located in the different underlying structures of the object-referencing morphemes in Inuktitut and Kalaallisut. Because Kalaallisut only has genuine φ-agreement, the approach adopted here predicts an absence of cooccurrence restrictions between pronouns and argument-referencing morphology.
6 Broader Discussion: Object Shift across Inuit
In the remainder of the article, I provide further theoretical context for the object φ-agreement vs. clitic-doubling distinction across Inuit. Why is it that Kalaallisut has object φ-agreement while Inuktitut has object clitic doubling, rather than the other way around? Moreover, why do the morphological properties seen in Inuktitut appear so idiosyncratic from the perspective of clitic doubling crosslinguistically?26
I propose that this contrast can be tied to a broader point of variation in object shift across the Eskimo-Aleut language family, akin to the pattern in Scandinavian (e.g., Holmberg 1986, Vikner 2006, Thráinsson 2008). In Icelandic, (27), object shift of DPs is optional and correlates with a semantic distinction similar to specificity (Diesing 1992, 1996); object shift of pronouns is obligatory, though omitted here for reasons of space. In contrast, Mainland Scandinavian languages such as Danish only permit pronominal object shift, (28).
Hann les sjaldanlengstu bókina.
he reads seldom longest the.book
‘He seldom reads the longest book.’
Reading: ‘Given any group of books, he seldom reads the longest one.’
Hann les lengstu bókinasjaldan.
he reads longest the.book seldom
‘He seldom reads the longest book.’
Reading: ‘There is a book longer than all the others that he seldom reads.’
Studenten læste denikke (*ikkeden).
student.the read it not
‘The student didn’t read it.’
Studenten læste ikkebogen (*bogenikke).
student.the read not book.the
‘The student didn’t read the book.’
As mentioned in section 4.1, the behavior of objects in Icelandic is parallel to that in Kalaallisut, though object movement in Kalaallisut is reflected by a difference in case morphology (ABS vs. MOD) rather than word order. Strikingly, Woolford (2017) draws an additional parallel between Mainland Scandinavian and Aleut, which is distantly related to the Inuit languages. In Aleut transitive sentences, the subject is always cross-referenced by φ-morphology, while the object normally is not, (29a). However, when the object is a pronoun, it appears on the verb within a portmanteau with subject φ-agreement, (29b).27
As Woolford points out, this is highly reminiscent of the agreement patterns in antipassive (ABS-MOD) and ergative (ERG-ABS) transitive sentences in Inuit. Thus, the Aleut examples can be analyzed as featuring an in-situ object DP in (29a), but a structurally high pronominal object in (29b)—just as in Mainland Scandinavian.
Here, I point out an additional crosslinguistic similarity. In certain Romance languages (e.g., Standard French), pronominal objects that obligatorily move are realized as verb-adjacent clitics, (30). I suggest that the Aleut example in (29b) displays an analogous effect.
Marie voit Jean.
‘Marie sees Jean.’
Marie le voit.
Marie him sees
‘Marie sees him.’
This triangulation between Mainland Scandinavian, Aleut, and French is reminiscent of proposals that have recast Mainland Scandinavian object shift as pronominal cliticization (e.g., Josefsson 1993, Bobaljik and Jonas 1996). As Holmberg and Platzack (1995) point out, however, there are several morphological reasons against adopting such an analysis wholesale. To resolve this (though I leave a fuller development for future work), I suggest that all three of these language groups display object shift of a pronominal D0—however, M-Merger applies only in Aleut and French, yielding a clitic.28
The preceding discussion provides crucial new insights into the question of why Kalaallisut and Inuktitut display a φ-agreement vs. clitic-doubling split. As demonstrated in (32), there is an inverse correlation between the relative “pronominality” of object agreement and the degree of object shift permitted. Moreover, Inuktitut occupies an intermediate position between Kalaallisut and Aleut along both dimensions. As in Kalaallisut, object-referencing morphology in Inuktitut may cross-reference a full ABS DP; however, as in Aleut, this morphology is a pronominal clitic. Furthermore, although D0s and DPs in Inuktitut may both undergo object shift, on par with Kalaallisut, the application of the operation Reduce (section 5.1) means that Inuktitut also patterns with Aleut in that the nominal element in AgrOP is structurally a D0.
Therefore, in addition to the independent evidence for object clitic doubling in Inuktitut, this section has provided broader context to the core empirical proposal of this article: that the object-referencing forms in Kalaallisut and Inuktitut are underlyingly distinct at a structural level, despite their morphological similarities.
In this article, I showed that genuine object φ-agreement and pronominal clitic doubling coexist within the Inuit languages. I demonstrated that Inuktitut displays a number of ABS object asymmetries and argued that this follows from an analysis in which subject-referencing morphology is genuine φ-agreement, while object-referencing morphology is actually clitic in nature: a pronominal D0. Furthermore, the Inuktitut data presented in this article are at odds with many previous characterizations of the Inuit languages—particularly Kalaallisut (e.g., Bittner 1994). Unlike Inuktitut, Kalaallisut exhibits no ABS object asymmetries, which suggests that it lacks clitic doubling; thus, in Kalaallisut, both subject- and object-referencing morphemes are realizations of genuine φ-agreement. Finally, I demonstrated that identifying these two divergent patterns in Inuit permits a more nuanced understanding of variation within Eskimo-Aleut morphosyntax.
Crucially, I arrived at this conclusion without referencing any morphological diagnostics for φ-agreement vs. clitic doubling. Recall that Inuktitut and Kalaallisut have nearly identical agreement forms, shown in (5) and (7), and that they uniformly fail standard morphological tests for clitichood (both showing mood variance). Examining their morphological appearance alone would have obscured the main empirical finding of this article: that Inuktitut and Kalaallisut object-referencing morphemes are underlyingly structurally different. This distinction instead emerged from examining interpretive and distributional interactions between these morphemes and the ABS objects they cross-reference.
More broadly, although there has been some work suggesting that all apparent instances of object-referencing morphology are doubled clitics (Woolford 2008, Nevins 2011), the analysis presented here suggests that such a treatment is too strong. Both are attested in natural language and may even coexist within a single language family.
1 In some languages, the clitic doubling of direct objects yields certain readings of the doubled DP that do not arise when other kinds of arguments (such as indirect objects) are clitic-doubled (e.g., Suñer 1988, Dobrovie-Sorin 1990, Bleam 2000). The clitic doubling of direct objects may also contrast with the clitic doubling of experiencers of psych predicates or raised possessors; whereas the former is usually optional, driven by information-structural considerations, the latter is obligatory in many languages (e.g., Kallulli 2000, Krapova and Cinque 2008, Harizanov 2014, Kramer 2014).
3 More precisely, Nevins (2011) proposes that, if a given argument-referencing morpheme displays tense variance, it must be genuine φ-agreement and not clitic doubling; if it is tense-invariant, then the diagnostic is inconclusive, not suggestive of either analytic option. In addition to tense (in)variance, Nevins proposes person complementarity (i.e., the Person-Case Constraint) and omnivorous number effects as diagnostics for clitic doubling. Neither of these effects exist in Inuit; hence, they cannot be tested. Moreover, Preminger (2011, 2014) has shown that omnivorous number effects can also arise in languages with genuine φ-agreement. Thus, in what follows I concentrate on Nevins’s morphological invariance diagnostic.
4 In the same vein, another purportedly morphosyntactic diagnostic for the φ-agreement vs. clitic-doubling distinction comes from the idea that pronominal clitics might be expected to resemble the independent pronouns of the language, if both are D0s (e.g., Preminger 2014). However, morphological similarity is still compatible with a φ-agreement account, as there is nothing to rule out exponing a φ-feature bundle with some form that resembles a pronoun with the same φ-features.
5 Similarly, if clitic doubling is derived by movement, then it should display effects of movement; for instance, it might be subject to intervention or be able to create new antecedents for anaphor binding (Preminger 2009, Anagnostopoulou 2003, 2016, Harizanov 2014). However, because these movement-based diagnostics are difficult to test in Inuit, due to the relative freedom of word order and some complicating factors concerning the distribution of anaphors (Bok-Bennema 1991, Beach 2011), I will set them aside in this article.
6 The unattributed Inuktitut data in this article were elicited in the community of Iqaluit, Nunavut, Canada, during three fieldwork trips in August 2016, August 2017, and October 2017.
7 In the data below, I do not gloss declarative mood/clause type, for simplicity.
8 The term modalis (MOD) is often used in the literature on Inuit to refer to the -mik/-mit-marked object in antipassive constructions, though other labels include ACC, OBL, or simply MIK case. Outside of antipassive objects, this case morpheme has a variety of functions, as it is also used to mark certain instrumentals, secondary predicates, and stranded modifiers of incorporated objects.
9 Thus, when they do encode the subject and object separately, as in (31b), only subject φ-agreement displays this allomorphy.
10 The declarative forms provided for Inuktitut and Kalaallisut differ slightly, because in Inuktitut the declarative is expressed using what is called the participial mood, while in Kalaallisut the relevant declarative-encoding mood is the indicative mood (Dorais 1988).
11 In contrast to Compton (2016), Johns (2017) and Johns and Kučerová (2017) argue for a clitic-doubling analysis of Inuit object-referencing morphology. Their evidence is mainly drawn from Eastern Canadian Inuktitut dialects (including the varieties presented here), though they generalize their proposal to all of Inuit. As I will show throughout the rest of this article, however, this generalization is incomplete given the contrast between Inuktitut and Kalaallisut.
13 In contrast, Wharram (2003) presents data indicating that these arguments take widest matrix scope, for example, scope out of embedded clauses and even islands. As movement cannot occur out of islands, Wharram argues that an account based on syntactic scope is insufficient, opting instead for a choice function analysis, building on work by Reinhart (1997), Kratzer (1998), Matthewson (1999), and others, though he also assumes movement of the object to Spec,TP, for reasons of case. Wharram’s data are problematic for the received view of the Inuit syntax-semantics interface as being movement-driven. As his findings have not been replicated elsewhere in the literature, I leave for future research the question of whether these approaches can be reconciled.
14 See Carrier 2016 for similar discussion along these lines, based on sociolinguistic evidence.
15 See Hallman 2008 for a similar observation.
16 Given an observation made by Wharram (2003) that ABS arguments in Inuit take widest scope even in islands, discussed in 13, a reviewer asks how that bears on the present diagnostic, which relies on the opacity of islands. The crucial point about (13) is the contrast between ABS subjects and ABS objects.
17 A reviewer points out that an effect similar to D-linking is also seen in other Romance languages like French when wh-objects cooccur with participle agreement (e.g., Obenauer 1994, Déprez 1998). As Tsakali and Anagnostopoulou (2008) observe, in languages that display both pronominal object clitics and participle agreement with objects (and object clitics), the two phenomena are in complementary distribution, meaning that object clitic doubling is impossible if participle agreement is present, and vice versa. This generalization strongly suggests a unifying factor behind these two sources of D-linking, though a more thorough investigation is beyond the scope of this article.
21 The Inuktitut data are in principle also compatible with a “big DP” treatment of object clitic doubling, whereby the pronominal D0 and its associated DP are generated as a constituent before the clitic undergoes long head movement to its final landing site (Torrego 1988, Uriagereka 1995, Arregi and Nevins 2012). However, as I discuss in section 6, it becomes difficult to reconcile this treatment with other Inuit varieties, such as Kalaallisut, that do not have object clitic doubling.
22 As Kramer (2014:618) discusses, it is perhaps puzzling that the object’s φ-features are not exponed on both AgrO0 and D0 if clitic doubling is Agree-driven. Kramer suggests that this could be due to a principle of morphological economy, coupled with a preference to expone the pronominal element over the agreement head (Rezac 2008, Preminger 2011). Another possibility I offer here concerns differences in the featural specification of the probe in AgrO0, such that it bears both φ- and movement-triggering features in Kalaallisut but only the latter in Inuktitut.
24 The NP-D0 word order presumably follows from Inuit’s head-final nature.
25 This inability to cooccur with independent pronouns is not specific to Inuktitut; it has also been observed as a property of clitics in dialects of Zapotec (Sichel and Toosarvandani 2018).
26 This question was raised by an anonymous reviewer. Although this section suggests that the answer lies in the broader variation in object shift across Eskimo-Aleut, note that a prediction made by this article is that these properties are not as rare as previously assumed. Rather, if most previous literature has relied on morphological diagnostics for φ-agreement vs. clitic doubling, then we would of course only find φ-agreement as affixes and pronominal clitics as morphophonological clitics. If the contributions of this article are on the right track, however, then we expect the reanalysis of previous data to yield different results.
27 This is accompanied by a change in case morphology on the transitive subject, termed the Aleut effect (e.g., Bergsland 1997, Sadock 2000). See Merchant 2011 and Yuan 2018 for recent analyses of this alternation. Interestingly, certain Inuit varieties spoken in Labrador, Canada, have been observed to display a pattern similar to that of Aleut (Johns 1999, 2001, 2017). Why two very distantly related Eskimo-Aleut grammars have converged in this way is outside of the scope of this article, but presents an intriguing puzzle for future research.
Thank you to Ragilee Attagootak, Susan Idlout, Shirley Kunuk, Jeanine Nowdluk, Jasmine and Jonah Oolayou, Jamesie Padluq, and two anonymous Inuktitut speakers for their discussion of the data included here and for sharing their language with me, and to the Nunavut Research Institute for logistical assistance. Thank you to Athulya Aravind, Nico Baier, Julien Carrier, Amy Rose Deal, Vera Gribanova, Heidi Harley, Sabine Iatridou, Alana Johns, Ivona Kučerová, David Pesetsky, Norvin Richards, Jerry Sadock, Ivy Sichel, Maziar Toosarvandani, Coppe van Urk, and two anonymous reviewers for helpful discussion and comments. This work was partially supported by a SSHRC doctoral fellowship and an NSF Doctoral Dissertation Research Improvement Grant. All errors are mine.
Abbreviations: 1PL = 1st person plural, 1SG = 1st person singular, 2SG = 2nd person singular, 3PL = 3rd person plural, 3SG = 3rd person singular, ABS = absolutive case, ACC = accusative, ADV = adverb, AGR = agreement, ALLAT = allative case, AP = antipassive, ASSOC = associative, AUX = auxiliary, CL = clitic, CONJ = conjunct, CONS = consequential, CTMP = contemporative mood, DECL = declarative mood, DEM.PRON = demonstrative pronoun, DUB = dubitative mood, ERG = ergative case, HAB = habitual, IND = indicative mood, INT = interrogative mood, ITER = iterative, LOC = locative case, MOD = modalis case, NEG = negation, O = object, PERF = perfective, PL = plural, PRES = present tense, PST = past tense, REC.PST = recent past tense, REL = relative case, S = subject, TNS = tense