Case as an Anaphor Agreement Effect: Evidence from Inuktitut

Since Rizzi 1990, it has been observed that anaphors across languages resist covariance with φ-agreement, a phenomenon now known as the anaphor agreement effect (AAE). In Italian, for instance, verbal φ-agreement cross-references nominative (NOM) arguments, which, while typically associated with subjects, may also surface on objects. While φ-agreement with a NOM object is normally licit, this becomes impossible if the object is an anaphor, (1a). That this ungrammaticality arises from covarying φ-agreement with the anaphor is further evidenced by (1b), as default (3SG) φ-agreement may ameliorate the sentence. Subsequent crosslinguistic work has shown that languages make use of a wide range of strategies, all conspiring to avoid covarying φ-agreement with anaphors (e.g., Woolford 1999, Haegeman 2004, Tucker 2011, Patel-Grosz 2014, Sundaresan 2014, 2016, Murugesan 2019, Preminger 2019, Rudnev 2020).

In this squib, I identify and confirm a prediction arising from two particular observations by Woolford (1999), thus also expanding our understanding of possible AAE strategies crosslinguistically. The empirical focus of the squib is an AAE strategy found in Inuktitut (Eastern Canadian Inuit).¹ On the basis of original fieldwork, I demonstrate that anaphors in Inuktitut are lexically specified as projecting additional syntactic structure (a K(ase)P, following Bittner and Hale (1996a) on related Inuit varieties), with the head of this projection realized as oblique (“modalis”; henceforth MOD) case morphology. Furthermore, I argue that this case layer on Inuktitut anaphors is obligatorily present in all syntactic contexts (regardless of whether it is morphologically realized). Because only ergative (ERG) and absolutive (ABS) arguments may be targeted by φ-agreement in Inuktitut, a φ-probe will fail to Agree (in the sense of Preminger 2011, 2014) whenever it encounters an oblique-marked anaphor. The basic case pattern and proposed structure are given in (2)–(3).

This exact AAE strategy is unique to Inuktitut and is previously unattested. However, its existence is simultaneously expected, given what we otherwise know about the broader typology of AAE strategies. As discussed by Woolford (1999), anaphors that may not appear in φ-agreeing positions may exceptionally do so if assigned quirky case; moreover, anaphors are commonly lexically specified as enclosed in larger structural constituents (e.g., possessive DPs). Although these have been previously treated as distinct, nonoverlapping strategies, I propose that the Inuktitut AAE is simply the intersection of these two patterns: Inuktitut anaphors are lexically specified as enclosed within a larger case-bearing structure. That the novel AAE strategy in Inuktitut arises directly from combining otherwise crosslinguistically familiar syntactic ingredients is a welcome finding, in line with standard generative tenets casting linguistic variation as shaped by common building blocks.

Finally, this analysis of Inuktitut departs from the received view that the Inuktitut AAE strategy involves detransitivization and that the oblique case on an anaphor in object position actually reflects its status as an adjunct (Marantz 1984, Bok-Bennema 1991, Woolford 1999). This view will be shown to be untenable: anaphoric objects in Inuktitut are true arguments of the verb, on par with their nonanaphoric counterparts.

This squib is organized as follows. Section 2 provides an overview of the AAE, focusing on the aforementioned observations by Woolford (1999). Section 3 provides several pieces of evidence that anaphors in Inuktitut are obligatorily and immediately dominated by a KP, whose head is exponed as oblique case morphology, and that this structural layer serves as an intervener for φ-Agree. Section 4 argues against a detransitivization-based approach to reflexivity in Inuktitut, building on previous work by Michael and Spreng (2014). Finally, section 5 concludes.

Besides the use of default agreement, shown in (1), this section highlights two other AAE strategies discussed by Woolford (1999): (a) the exceptional ability for anaphors to surface as quirky (lexical casemarked) subjects in languages like Icelandic, and (b) the crosslinguistic containment of anaphors within complex DP structures, exemplified here by Selayarese. As I will show, the AAE strategy found in Inuktitut combines these strategies in a novel way.

Like φ-agreement in Italian, φ-agreement in Icelandic typically only targets NOM arguments. Encountering a non-NOM DP thus results in failed Agree in the sense of Preminger 2011, 2014, in that the probe is simply left unvalued (and exponed as default 3SG agreement). This logic of case/agreement interactions also underlies the nature of the AAE in Icelandic, in that anaphors may not be NOM. This is illustrated below with the anaphor SIG,² which may be bound long-distance as a subject across a subjunctive clause boundary. Crucially, while binding of SIG in subject position is not possible if it bears NOM case, (4a), it becomes available when SIG receives quirky (e.g., DAT) case, (4b) (Maling 1984). As discussed by Rizzi (1990) and Woolford (1999), this contrast cannot be attributed to a morphological gap in the anaphor’s paradigm; rather, it is due to the AAE.

The second relevant AAE property highlighted by Woolford (1999) pertains to the crosslinguistic tendency for anaphors to be enclosed in possessive and body-part DPs (Faltz 1977, Schladt 2000). This is illustrated with Selayarese in (5). At first glance, Selayarese displays φ-agreement with anaphoric objects. However, this φ-morphology is invariably 3SG, regardless of the featural specifications of the anaphor.³ Woolford proposes that this is in fact agreement with the complex DP, not the anaphor within the DP. This additional structure serves as an intervener for φ-Agree, preventing the φ-probe from accessing the anaphor internal to this structural material.⁴

We have now seen two manifestations of the AAE, wherein anaphors may be enclosed within agreement-repelling syntactic structure. These strategies seem descriptively similar, aside from the category of the relevant structural layer (KP vs. DP). However, they are in fact fundamentally distinct in a deeper way. In the Icelandic-type pattern, this structural layer is not inherent to the anaphor; when it is present, it is due to independent syntactic factors relating to argument structure or idiosyncratic case-assignment properties of certain verbs. In contrast, in the Selayarese-type pattern the structural layer is part of the lexical specification of the anaphor itself, in that all anaphors in the language are enclosed within a larger DP constituent, regardless of other aspects of the sentence-level syntax.

Turning now to Inuktitut, I demonstrate that it displays a third kind of AAE pattern that combines the core ingredients of these strategies: in Inuktitut, anaphors are lexically specified to be contained within a KP layer, realized as oblique case morphology. While the interaction between case and the AAE in Inuktitut is seemingly reminiscent of the Icelandic quirky case pattern, the crucial difference is that the oblique case in Inuktitut is insensitive to syntactic context, since it is a core property of the anaphor itself. As we will see in section 3, the case layer is obligatorily present on anaphors in a variety of syntactic positions, even when the anaphor is not an argument of the verb. This suggests an analysis of Inuktitut parallel to that of Selayarese—only the additional structure in Inuktitut is an oblique KP rather than a possessive DP. The commonalities and differences between the Icelandic, Selayarese, and Inuktitut patterns are thus summarized in (6).

In Inuktitut, φ-agreement may target ABS-and ERG-marked nominals (Bobaljik 2008), while nominals bearing other cases (e.g., oblique cases such as MOD -mik in the antipassive example in (7b)) may not be indexed by agreement morphology.⁵

Inuktitut has several oblique cases, given in (8). Most of these cases display contextual allomorphy, with the choice of morph determined by the grammatical properties of the stem to which it attaches (see Yuan 2015 for details). For present purposes, the anaphor immi surfaces with the variant on the right.

We have already seen in (2b) that anaphors in Inuktitut may surface with MOD case. I now argue that this is a lexical property of the anaphor, due to the AAE: anaphors in Inuktitut are lexically specified as immediately dominated by a K(ase)P-layer, whose head is realized as MOD case morphology, (9). Because only ERG and ABS nominals are able to be targeted by φ-Agree, anaphors may never be cross-referenced by φ-agreement.

The analysis of Inuktitut anaphors is largely informed by comparing the distribution of MOD case on anaphors (henceforth MOD_ANAPH ) with that of the other oblique cases, including MOD case on antipassive objects (henceforth MOD_AP).⁶ Despite their morphological identity, I show that these instances of MOD case have distinct structural sources.

The first piece of evidence that anaphors are lexically specified as bearing MOD case comes from the novel observation that MOD case surfaces even in contexts in which case otherwise cannot be assigned to nonanaphoric nominals. One such context is within ‘picture of’ complex DPs, which may be expressed in Inuktitut with the nominal ajjinnguaq ‘picture’ modified by another DP-internal nominal.⁷ Both nominals bear the case that is assigned to the complex DP as a whole, which I assume is the result of a case concord process. This is first shown in (10a–b), with nonanaphoric objects.

Crucially, when the modifier of ajjinnguaq is an anaphor, it is obligatorily marked with MOD_ANAPH case morphology, (11); in (11b–c), we additionally see case stacking, as the anaphor surfaces with both MOD_ANAPH and the case assigned to the entire DP. Finally, (11c) demonstrates that case stacking persists even in the absence of an intervening modifier. Note that, in these examples, the final /k/ of the MOD case form -nik does not surface due to a regular morphophonological rule (Dorais 1986).

Throughout (11), the morpheme order is ANAPHOR-MOD_ANAPH-ADJ-CASE2. Assuming that morpheme order reflects syntactic hierarchy (Baker 1985), this means that the anaphor is immediately dominated by the MOD_ANAPH KP, which may then be dominated by an AdjP if one is present; this complex, in turn, bears the case assigned to the entire DP (“CASE2”).⁸ The MOD_ANAPH > ADJ order on anaphors is inviolable, as shown in (12a), although CASE > ADJ orderings are otherwise generally impossible on nonanaphors, (12b).

An apparent exception to the above comes from high adjectival suffixes such as -tuaq ‘only’ (c-modifiers in the terminology of Cardinaletti and Starke 1999), which participate in both CASE > ADJ and ADJ > CASE morpheme orders when attached to nonanaphors, with no apparent difference in meaning, (13). Nonetheless, in accordance with our previous empirical generalization, the morpheme order is still rigidly CASE > ADJ on anaphors, (14).

To sum up, the idea that Inuktitut anaphors are lexically specified as oblique translates syntactically into anaphors being obligatorily and immediately dominated by a KP-layer, as in (9). This accounts for the syntactic contexts in which only anaphors (and not other DPs) may bear case, as well as the inability for any other syntactic elements to intervene between the two.

At this point, alternative analyses are still available: perhaps the sequence imminik is monomorphemic, or perhaps -nik is homophonous with MOD case but is a distinct morpheme. However, evidence that -nik on anaphors is indeed case morphology comes from systematic interactions with other oblique cases in the language. As shown in (15), the sequence -nik is absent when the anaphor is found in other oblique contexts. This is unexpected under the aforementioned alternative analyses.

I propose that this pattern is due to a haplology rule operating on structurally adjacent case morphemes (K⁰s), resulting in the deletion of the internal case morpheme (MOD_ANAPH), as stated and represented in (16a–b).

That this rule is resolved by deleting the inner case in particular, rather than the outer one, is unsurprising, given their grammatical functions. The (outer) oblique cases (K⁰₁) in Inuktitut contribute semantic information, often encoding adositional (e.g., spatial or relational) information; this information would be lost if the haplology rule applied to K⁰₁. In contrast, if the present analysis is correct, the (inner) MOD_ANAPH case (K⁰₂) on anaphors serves only to circumvent the AAE but is otherwise semantically vacuous, so may be a better (more expendable) candidate for deletion. This contrast is in line with a broader tendency in case-stacking languages to prioritize realizing semantically contentful cases over nonsemantic (e.g., structural) ones, as discussed by Richards (2013) on the basis of Lardil.⁹

MOD_ANAPH thus surfaces whenever the environment triggering the haplology rule is not met. This is further evidenced below. Recall from (13)–(14) that the adjectival suffix -tuaq ‘only’ may optionally attach higher or lower than a (non-MOD_ANAPH) case layer. Crucially, this affects whether MOD_ANAPH surfaces on the anaphor. If the adjective is merged above the outer K⁰ (such that the two K⁰s are structurally adjacent), the rule in (16) applies and MOD_ANAPH does not surface, (17a); if it intervenes between the two K⁰s, then (16) does not apply and case stacking results, (17b).

Finally, I return to the pattern discussed in section 3.1—that is, the appearance of case stacking in examples like (11c) despite the absence of an intervening adjective. I suggest that the crucial distinction between case haplology and case stacking boils down to whether or not the multiple case layers are associated with the same DP constituent. In examples such as (15), the K⁰s are structurally contiguous and both associated with the anaphor; in this context, haplology applies, deleting the internal case morpheme. However, in case-stacking examples such as (11c), the outer oblique case is assigned to the complex DP properly containing the anaphor, rather than the anaphor itself, but is morphologically realized on the anaphor due to nominal concord.¹⁰ Here, the two K⁰s are not structurally adjacent, and thus the haplology rule in (16) does not apply.

I now illustrate how the proposed oblique KP-layer interacts with φ-Agree in order to circumvent the AAE, without affecting the broader argument structure. As already seen in (2), anaphoric objects of transitive verbs trigger an ABS-MOD case frame, with φ-agreement indexing only the subject. That this is the result of failed Agree is most straightforwardly demonstrated with predicates bearing the transitivizer -gi. This morpheme introduces an external argument and embeds otherwise intransitive predicates such as psych-predicates (as in (2)) and certain noun-incorporating constructions, (18). I analyze this morpheme as a v⁰ and the argument as its specifier; its complement in (18b) is the noun-incorporating verb phrase predicate, suggesting a structure as in (19).¹¹

Crucially, -gi-transitivized verbs may not be antipassivized—meaning that an ABS-MOD case frame is normally not ever possible, as shown by the ill-formedness of (20a).¹² However, (20b) demonstrates that the otherwise impossible ABS-MOD case pattern exceptionally surfaces when the object is an anaphor (see also (2b)). Note also that the loss of ERG case on the subject is suggestive of a dependent treatment, in that the presence of MOD_ANAPH on the anaphor removes the case competitor for dependent case assignment to the subject (Marantz 1991, Baker 2015) (for independent evidence that ERG in the Inuit-Yupik-Unangan language family is a dependent case, or assigned configurationally, see Bittner and Hale 1996a,b, Baker and Bobaljik 2017, and Yuan 2018).

The pattern in (20b) is unsurprising given the proposed structure of anaphors. MOD case morphology on the anaphor arises from its lexically specified KP-layer, not antipassivization. Moreover, since such KPs are syntactically opaque, the φ-probe will inevitably fail to be valued—resulting in the absence of object φ-morphology. Finally, the very fact that this pattern is, again, specific to anaphoric objects converges with the previous characterization of MOD_ANAPH from sections 3.1–3.2, thus providing further evidence for the exact AAE strategy proposed in this squib.

The analysis of Inuktitut anaphors presented here departs from the more received view that such constructions involve detransitivization (Marantz 1984, Bok-Bennema 1991, Woolford 1999), in line with arity-reducing (i.e., detransitivizing) approaches to reflexivization (e.g., Reinhart and Siloni 2005). Under this view, the intransitive predicate by itself is sufficient to yield a reflexivized reading, oblique case morphology on the anaphor being due to its status as an adjunct (this also accounts for the loss of ERG case on the subject). However, the data above have already suggested that this approach is insufficient—for instance, it does not explain the distribution of MOD case on anaphors in complex DPs (section 3.1), nor its interaction with other oblique cases (section 3.2).

There is, however, an additional, more crucial piece of evidence against such an approach: anaphors cannot be omitted in Inuktitut, a fact not expected of adjuncts. As first observed by Michael and Spreng (2014), omitting the anaphor either eliminates the reflexive reading or renders the sentence ungrammatical altogether, as demonstrated in (21).¹³ Michael and Spreng additionally show that this is a point of variation across Inuit; the anaphor does seem to be optional in other varieties such as Kalaallisut and Iñupiaq, (22).

Thus, a detransitivization-based analysis of reflexive constructions is untenable for Inuktitut, even though it may be correct for other Inuit varieties; the MOD-marked anaphoric objects of transitive verbs in Inuktitut are true arguments.

In this squib, I have demonstrated that anaphors in Inuktitut are lexically specified as enclosed within a KP, such that they obligatorily bear oblique case morphology. This is an anaphor agreement effect: because obliques cannot be targeted by φ-Agree processes, a φ-probe that encounters an anaphor will inevitably fail to be valued. Though previously unattested, this pattern is a welcome addition to the existing typology of AAE strategies, given its structural parallels with previously observed patterns (Woolford 1999).

Although it is beyond the scope of this squib to explain why the AAE holds (in Inuktitut and in general), the Inuktitut pattern offers a novel explanandum for existing theories. For instance, it is incompatible with the account advanced by Preminger (2019) according to which anaphors are universally composed of a φ-bearing core contained within a structural layer (“AnaphP”) that both contributes the nominal’s anaphoricity and prevents φ-Agree by a higher probe. While this seems analogous to the structural approach taken here, the relevant opaque structure in Inuktitut is clearly a case layer (see section 3.2) and is thus external to the anaphor, regardless of the anaphor’s internal composition. At the same time, the Inuktitut data present a conceptual challenge for approaches that connect the AAE to the idea that anaphors lack φ-features altogether and therefore cannot value a φ-probe (e.g., Shiraki 2004, Murugesan 2019). If the Inuktitut pattern involves failed Agree, and φ-probes may simply be left unvalued, such approaches fail to explain why an intervening structural layer above the anaphor is needed at all.

Thank you to Shirley Kunnuk, Jasmine Oolayou, and especially Ragilee Attagootak for their insights on the data included here and for sharing their language with me, and to the Nunavut Research Institute for logistical assistance. Thank you to Nico Baier, Sabine Iatridou, Alana Johns, David Pesetsky, Norvin Richards, and Sandhya Sundaresan for helpful discussion, as well as audiences at MIT, UQAM, NELS 49 (Cornell University), and DISCO (Universität Leipzig). This work was partially supported by a SSHRC doctoral fellowship and an NSF Doctoral Dissertation Research Improvement Grant (BCS-1728970). All errors are mine.