1 Introduction

Multiple Exponence (ME), in the explicit definition of Harris (2017: 9) is “the occurrence of multiple realizations of a single morphosemantic feature, bundle of features, or derivational category within a word”. Assembling an extensive array of case studies, Harris (2017) showed the phenomenon to be fairly common cross-linguistically. As the examples in (1) illustrate, a wide variety of features can be multiply exponed.

ME does not constitute a uniform phenomenon. To address different instances of ME, several different mechanisms have been proposed in the Distributed Morphology literature so far. Some ME phenomena have been shown to be derived by idiosyncratic, but well-motivated postsyntactic operations: Fission (Noyer 1997; Arregi & Nevins 2012; McGinnis-Archibald 2016); phonological reduplication of the relevant exponents, (Harris & Halle 2005; Arregi & Nevins 2012; Deal 2016), and “enrichment,” which essentially is a reduplication of morphological features, Müller (2007). On the other hand, some ME phenomena have been shown to arise still in the narrow syntax and to be a manifestation of a more involved syntactic structure, but not to require any non-standard syntactic operations (Baker & Willie 2010; Oxford 2017).

In this paper, I address an apparently typologically rare phenomenon, multiple case exponence in the plural forms of wh-words and wh-based indefinites in Digor Ossetic, an endangered minority East Iranian language spoken in the North Caucasus. The pattern of case marking in such items is illustrated in (2) with a partial paradigm of the wh-word ka ‘who’ and the indefinite ka-dɐr who-idf ‘someone’. In all the cases with non-null exponents, i.e., all the cases other than the nominative, case marking appears twice: after the wh-stem and after the plural marker.

I will argue that this instance of ME arises due to a combination of a rather idiosyncratic internal syntax of these items and a postsyntactic operation motivated by haplology avoidance. More specifically, I propose that such plural indefinites involve two D heads, each of which receives the case feature. The higher D head receives the feature by the regular case-assignment mechanism, while the lower one receives the same feature from the higher by a dedicated last-resort feature-sharing mechanism. A similar feature-sharing procedure was proposed in Clem & Dawson (2021: 13) for multiple occurrences of D in nominals in Tiwa (Tanoan). I provide additional language-specific evidence for the existence of this mechanism in Digor Ossetic and propose a locality condition for it. At the spellout, a haplology-resolving operation takes place under certain conditions in Digor, which results in the spellout of non-identical case markers.

This paper is organized as follows. In Section 2, I introduce the empirical puzzle this paper deals with, double case exponence in certain wh-based plurals. In Section 3, I show that postsyntactic operations proposed in the DM literature to account for Multiple Exponence cannot account for the phenomenon under discussion. Section 4 provides the necessary background on Digor Ossetic case and number morphology, and on the structure of wh-based indefinites in this language. Section 5 systematically addresses case and number marking in wh-based items. Section 6 treats the syntax of DPs in Digor in general, and the internal structure of wh-items of different types, wh-based indefinites, and of their plural forms. It introduces the first crucial ingredient of the analysis – the distinction between nP-sized and DP-sized wh-words, and the presence of two copies of D in the plural forms of the latter. Section 7 uses the syntax developed in Section 6 to show that narrow syntax-based proposals advanced for other instances of ME by Baker & Willie (2010) and Oxford (2017) cannot be generalized to our situation. Section 8 deals with case assignment in nominals involving two D heads. It introduces the second crucial technical ingredient of the analysis – the operation of last-resort feature transmission. Section 9 addresses haplological dissimilation of case exponents. Section 10 discusses possible reasons for the cross-linguistic rarity of the Ossetic-type double case marking pattern.

Unless explicitly indicated otherwise, all examples in this paper are from Digor Ossetic. Unattributed examples all come from my fieldwork materials.

2 The Puzzle

The phenomenon to be addressed in this paper is a typologically unusual pattern of multiple case exponence in the plural forms of some wh-words and wh-based indefinites in Ossetic. This phenomenon has not been previously addressed in the theoretical literature. In more traditional literature, Zgusta (1965) discussed the data and noticed their typological uncommonness.²

Ossetic builds indefinites on the basis of wh-stems, which by itself is common cross-linguistically, see Haspelmath (1997) for a typological overview, and specifically Haspelmath (1997: 281) for an inventory of indefinites in Iron Ossetic and an overview of their functions. I will call the series of indefinites of interest to us the jes-series and dɐr-series, respectively.

The case marking in these indefinites is borne by the wh-stem, as illustrated by the partial paradigms of ‘who’-based indefinites in (3). For background information on Ossetic case morphology and complete paradigms, see Section 4 below.

Furthermore, wh-words and indefinites of jes- and dɐr-series allow plural marking (4). For wh-words, this fact by itself is not typologically unusual, at least not for the region where the Ossetic languages are spoken, see e.g. Kornfilt (1997: 316–317) for plural-marked wh-words in Turkish (Turkic, Turkey); Tschenkéli (1958: 196) for Georgian (South Caucasian; Georgia); Dum-Tragut (2009: 148) for Standard Eastern Armenian (Indo-European; Armenia); and Bagirokova et al. (2022: 289) for Adyghe (Northwest Caucasian; Russia).

While the singular forms are in principle number-neutral, the plural forms can be used when some kind of a plurality presupposition is involved (5).

Unexpectedly, in all the cases other than the nominative, the plural forms of these indefinites carry two overt case markers. One of them is the ‘actual’ case exponent on the wh-stem, and the other follows the plural marker, as is illustrated in (6) by a partial paradigm of ‘who’-based indefinites. For the plural forms of wh-words and jes-series indefinites, the outer case suffix is always the oblique case marker -i, (6 a–c). On the other hand, for the plural forms of the dɐr-series it can be either the obl marker -i or a second instance of the actual case marker (6 d–f).

The sentential examples in (7) below illustrate the use of these items.

It goes without saying that in the plural forms of lexical DPs, case is only marked once. Furthermore, no overt concord in case or number exists in Ossetic DPs (8b–d).

It is worth stressing that although the items in (6) and (7) bear two different case markers, that of the respective morphological case X and the oblique -i, the item as a whole only bears a single case normally expressed by X. This is illustrated by the minimal pair in (9), of which the sentence in (9a) repeats (7d). The lexical DP in (9b) only bears the superessive case marker -bɐl.

Therefore, what we are dealing here with is indeed an instance of multiple case exponence in the sense of Harris (2017: 9), and not Suffixaufnahme in the sense of Plank (1995). i.e., not the marking of a nominal with several case markers coming from several case assigners. For instance, in the Old Georgian example in (10), the noun iak’ob Jacob bears the genitive as the possessor of saxl ‘house’, and the ergative that is assigned to the entire DP ‘Jacob’s house’, possibly copied to the possessor by a case concord mechanism.

To account for the pattern in (6), I will argue that the structure of the respective plural-marked wh-words and wh-based indefinites involves two D heads, each of which is assigned case. The case value is shared between them, similarly to the proposal of Clem & Dawson (2021) for Tiwa. To express the case, each of the D heads projects an Agr head of its own. If those latter heads are close enough to each other, their features undergo dissimilation to avoid haplology, Nevins (2012).

3 Against Post-Syntactic Derivation of ME in Ossetic

In this section, I show that post-syntactic operations proposed in the literature to derive various manifestations of ME, cannot explain the Ossetic facts under discussion no matter what the syntax of the items involved. Specifically, I address fission, phonological reduplication, and enrichment.

4 Background on Digor Ossetic morphology

Digor and Iron Ossetic are closely related East Iranian languages spoken in the Central Caucasus. They are often called dialects of the same language in the literature. The data in this paper is from Digor, however, a similar phenomenon exists in Iron, and the analysis proposed here is applicable to the Iron facts as well.

5 Case marking of wh-words, wh-based indefinites, and their plural forms

In this section, I systematically lay out the morphological facts to be accounted for in this paper. Digor wh-words and indefinites derived from them divide into two classes: those that exhibit the double case marking sketched in Section 2, and those that behave like lexical nouns, i.e., ones that expone the case only once.

6 The structure of indefinites and of their plural forms

To proceed, we need to investigate the internal structure of indefinites and their plural-marked forms in Ossetic. To that end, and in order to explain the contrast in case exponence between lexical plurals and plural indefinites, we need to establish the overall structure of the DP in Ossetic.

7 Narrow-syntax-based analyses of ME

With the internal structure of the items under discussions at hand, we are now in the position to determine whether our phenomenon is amenable to the existing analyses that treat some instances of ME as a direct outcome of processes taking place in narrow syntax. I argue that neither of two such proposals is applicable to our situations.

Baker and Willie (2010) analyze multiple exponence of agreement in Ibibio. They build upon the proposal of Pesetsky & Torrego (2007) to argue that, in Ibibio, the (multiple) goals that have to agree in phi-features with the subject first form a feature sharing chain between themselves, and then, when the highest functional head participating in the dependency agrees with the subject in its specifier, all the lower goals participating in the chain have their features checked.

However, in our case, D heads presumably serve as agreement goals, and by default do not enter in the Agree relationship with each other.

Oxford (2017) analyzes the multiple exponence of verb agreement markers in some Algonquian languages. He uses the well-established fact that in Algonquian languages, the verb includes T and C heads, both of which may, or have to, undergo agreement with the subject. He proposes that, in a given language, certain goals may be able to agree with a probe which already underwent agreement with a different goal. This refines a proposal of Baker (2008), who had conjectured that such an ability (the Activity Condition) is a macroparameter, i.e. has to hold (or not to hold) for all the goals in a given language. On the proposal of Oxford (2017), in the Algonquian languages, if the respective T and C heads are able to agree with the same goal, double exponence emerges.

The analysis of Oxford (2017) cannot be extended to our case for the same reason as that of Baker & Willie (2010) cannot. The two D heads in the spine of our nominals (28) provide two goals, rather than two probes. For locality reasons, a probe will only agree with the nearest goal, that is, the higher D head. Accordingly, the valuation of the case feature on the lower D head will be blocked.

8 Case Assignment and Feature Transmission in structures with two D heads

I adopt the standard assumption that D heads are merged with their unvalued uninterpretable case feature, and for the derivation to converge, this feature must be valued. Accordingly, given the findings of Section 6.4, the plural form of an indefinite formed from a D-sized wh-word (i.e. of ‘who’ or ‘what’), or the plural form of such a wh-word itself, has two D heads with unvalued case features. I illustrate it in (29) for the plural of a dɐr-series indefinite.

I remain agnostic as to the mechanism by which a DP as a whole gets its case valued. It can be achieved by agreement with an appropriate functional head, Chomsky (1981; 2000; 2001), or by a case-assigning algorithm (“dependent case theory”), Yip et al. (1987), Marantz (1991), Levin & Preminger (2015), Levin (2017), or perhaps by both, Baker & Vinokurova (2010). The resulting morphology does not depend on it. What is crucial, however, is that no matter what mechanism of case assignment is at work, only the case of the higher D head will be valued, given the standard locality assumptions. The case feature of the lower D will remain unvalued, and that would prevent the derivation from converging.

Accordingly, a mechanism must exist that values the case feature of the lower D. The logically possible ways to achieve this are (A) to make the same mechanism assign it to the lower D as to the higher D, (B) to assign the default case to it; and (C) to copy the feature values from the higher D as a last-resort operation.

Option (A) clearly violates the standard locality assumptions. Option (B) is empirically inadequate, as it predicts that the case value on the lower D will not depend on the case assigned to the entire DP, which is patently false – it is precisely the lower, i.e. the innermost, D that exhibits the case assigned to the DP as a whole, e.g. the dative and the allative, respectively, in (31a–b).

Accordingly, we must make recourse to option (C), last-resort feature copying. A similar proposal has been made for the layered DP in Tiwa by Clem & Dawson (2021).

Independent evidence in favor of case transmission by feature copying between D heads comes from the case marking of appositives with pronouns, i.e., expressions of the type we linguists. In such appositives, both parts must be case-marked (32).

Such appositives form single DPs in Ossetic, as can be seen from their interaction with Wackernagel enclitics. DPs in Digor are impenetrable for 2P clitics, which otherwise are placed after the first word of the clause, (Erschler 2010: 7). Accordingly, impenetrability for clitics can be used as a DP-hood test. This is what obtains for appositives.

Accordingly, the structure of such an appositive must be as shown in (34), cf. a similar structure proposed for multiple definiteness marking in Greek in (Lekakou & Szendrői 2012: 114) and a proposal with a more fine-grained syntactic structure of determiners in (Kyriakaki 2020: 126).¹⁸

The case-marking facts in (32) then receive a natural interpretation if one makes use of the feature transmission operation.

The procedure of feature transmission proposed here might look suspiciously similar to concord, which, as we have seen in (17), is absent in Ossetic. However, the proposal does not overgenerate, because only DPs are taken to be able to receive case and to sprout the respective Agr nodes. Modifiers within a DP do not contain D heads and accordingly cannot participate in feature transmission.

As a locality condition on this kind of feature transmission, I assume that it cannot cross category-defining heads. This is consistent with the proposal of Marantz (2007) that such heads are phasal. Evidence in favor of this locality condition comes from the case marking of nominals involving several category-defining heads. The Ossetic languages cannot form denominal verbs synthetically, so manifestations of locality with category-defining heads other than n⁰ are hard to come by. There is no reason, however, to assume that last-resort feature transmission is an Ossetic-specific operation, so it is legitimate to look for such examples from other languages.

As one piece of evidence, consider the behavior of the German verbs duz-en ‘to address someone by Du’ and siez-en ‘to address someone by Sie’, which are formed on the basis of the respective pronouns. I assume German personal pronouns to instantiate, or at least to include, D⁰. These depronominal verbs can be nominalized in their turn: (das) Du-z-en ‘addressing by Du’, (das) Sie-z-en ‘addressing by Sie’. As nouns, these can be assigned case. However, the case cannot percolate to the embedded pronoun.

The absence of case percolation into the nominalisation is exactly what the proposed condition predicts. A category changing head intervenes between the two instances of D, as shown in (37), and case transmission becomes impossible. I assume that the embedded pronoun Du receives the default case.

Furthermore, the German das Ich ‘ego’ must involve an n⁰ layer atop of the DP corresponding to ich ‘I’. This is necessary to ensure that it has a noun-like distribution and be able to host gender, see Kramer (2015: 37) for the latter point. Given that the n-layer is impenetrable for case transmission, the noun becomes indeclinable, unlike the underlying pronoun. The case inflection patterns of the pronoun and the derived noun are compared in (38).

Similarly, the Russian depronominal noun ja ‘true self’, from ja ‘I’, is indeclinable, unlike the underlying pronoun. Again, the structure of this noun presumably is [_DP D⁰ [_nP n⁰ [_DP ja]]].

To recapitulate the discussion of this section, I have argued in favor of the existence of a last-resort operation, feature transmission, that, modulo appropriate locality conditions, allows a lower head with an unvalued feature F to have the feature valued against a higher head of the same type. In our case, the heads are D. The existence of this operation will explain the double exponence of case on plural indefinites. What remains unexplained so far, is the fact that the outer case marker can be the oblique instead of that of the expected specific case value.

9 Dissimilation of case exponents

In this section, I address the spellout process that results in non-identical case exponents appearing in the plurals. This is obligatory for bare wh-words and jes-series indefinites, where only the plural marker -tɐ- intervenes between the two case markers, and optional for dɐr-series indefinites, where two morphs, the plural -tɐ- and the indefinite -dɐr- intervene between them, as illustrated in (39).

I argue that the oblique case marker is inserted here to avoid haplology between the two case exponents. I propose that the oblique marker -i serves as the default non-nominative case marker. More specific exponents are prevented from being inserted by means of feature deletion. The difference between bare wh-words and jes-indefinites, on the one hand, and -dɐr-indefinites on the other hand, lays in the fact that -dɐr is a suffix, which creates a longer distance between the two case exponents. That allows case dissimilation in dɐr-series indefinites to be optional.²⁰

To lay out the analysis in full detail, consider the tree for a plural indefinite with the Agr nodes corresponding to the case markers (40). For the sake of concreteness, I show the tree for wh-dɐr-pl, the picture for the wh-pl and jes-wh-pl is fully identical, up to the position of the adjoint indefinite marker (28).

A priori, the following causes may lead to the non-identity of the exponents of Case₁ and Case₂ nodes. First, the case features assigned to D₁ and D₂ in narrow syntax may be non-identical. Otherwise, some dissimilation process must be implicated that leads to the spellout of the oblique in Case₂. This dissimilation might be purely phonological, i.e. it might be driven by the need to prevent identical syllable rhymes from appearing too close to each other. The remaining alternative is that dissimilation occurs at some stage of morphological derivation.

I will first consider and reject the possibility that the non-identity of the exponents of Case₁ and Case₂ is determined in narrow syntax or in late-stage phonology, and then will make use of the approach of Nevins (2012) to dissimilation in morphology.

The possibility that D₁ and D₂ are assigned non-identical features in syntax can be rejected outright. Indeed, case is assigned in syntax to the entire nominal, i.e. to DP₂. On the other hand, this case value is expressed on the inner head, D₁ as is illustrated in (39). Accordingly, what we are dealing with here is some kind of dissimilation.

Furthermore, the dissimilation cannot be merely phonological. First of all, it is highly implausible that a purely phonological process exists that would create /i/ out of a diverse assortment of segmental sequences, which the Digor case markers are, as was shown in Table 1 above. Second, linearly close sequences of segments that are identical to the case markers are in principle tolerated, as illustrated in (41) for the dative -ɐn and the ablative -ɐj.

Therefore, we must conclude that dissimilation occurs in morphology. Now, as Nevins (2012: 87–88) has argued, dissimilation of non-adjacent exponents must be brought about by feature deletion.²¹ Observe that, in our settings, the dissimilating case markers are separated at least by the plural marker, and, in dɐr-series indefinites, also by the indefinite marker (39). Accordingly, the items undergoing dissimilation are non-adjacent, and some feature deletion must occur.

To provide an appropriate feature representation of the case markers in Digor, I will, first of all, adopt the proposal to distinguish the abstract case and the morphological case that gets exponed, following McFadden (2004: 10–13); Legate (2008: 95); and the ensuing literature. In view of this, I assume that nominative forms lack morphological case features altogether. Furthermore, it is natural to consider the exponent /i/ as the default (non-nominative) case exponent, because it fulfills the functions of the accusative, the genitive and the inessive marker in the majority of the paradigms. For the rest of the cases, I assume that the featural representation is [nnom, case], with case running over dat, abl, etc.

Accordingly, with both agr nodes in (40) having the featural representation [nnom, x], the case-specific feature x is then deleted from the outer case feature bundle, and the vocabulary item (VI) with the representation [NNom], that is, /i/, is inserted.

As for the directionality of dissimilation, it is not clear that a principled explanation is possible of why it is the outer case node that undergoes feature deletion. As Nevins (2012: 88) notes, cross-linguistic variation obtains in this respect. If dissimilation was obtained by allomorph choice rather than by feature deletion, and if Vocabulary Insertion proceeded from the bottom to the top of the tree simultaneously with, or prior to, linearization (Bobaljik 2000; Embick 2010: 42; Schreiner 2021: 19–20), the fact that it is the outer exponent that dissimilates would receive a natural explanation. However, as we have seen, the more plausible dissimilation mechanism here is feature deletion, and, furthermore, a number of counterarguments have been raised against pre-linearization VI insertion (Arregi & Nevins 2012: 238; Merchant 2015: 276–281; Haugen & Siddiqi 2016: 369). Accordingly, the direction of dissimilation has to be taken to be a language-specific, or perhaps even language and feature-specific, parameter.

10 Cross-linguistic picture

Plural-marked wh-words are not particularly uncommon cross-linguistically. However, the Ossetic-type double case marking system appears to be rarely if at all attested elsewhere. Typically, the plural forms of wh-words only bear one case marker (42). Indeed, as we have seen, even within Digor, not all wh-words exhibit double marking. The analysis proposed in this paper allows us to hypothesize why such a system is so rare.

Indeed, the analysis developed here predicts that for a wh-item to exhibit double case exponence in the plural, the following very specific combination of circumstances needs to obtain. First, the structure must contain two D heads on the spine of a single M-word. Second, the lower instance of D within this item has to be unable to be spelled out with a default case value, but rather must have its case properly valued. Third, even if these conditions are fulfilled, one of the resulting case exponents should not be deleted to avoid haplology. Probably, even more processes may conspire to conceal the double case marking on the PF. It is not particularly surprising therefore that systematic double case exponence of the kind attested in Ossetic is a typological rarity.

An additional reason for the rarity of such systems may be diachronic, see Harris (2008; 2010) for a discussion of how diachronic processes can lead to the emergence of typologically rare structures. Specifically in the case under discussion in this paper, an additional factor might be at play. Haspelmath (1993) proposed that a diachronic tendency exists that he called “externalization of inflection”. If a situation arises when inflectional markers appear word-internally, it is diachronically unstable, and the inflectional markers tend to “migrate” to the edge of the word. This tendency, if real, would also contribute to the rarity of the double case marking pattern discussed here – it is then predicted to be diachronically unstable,²² see the general observation about diachronic instability as a factor contributing to the typological rarity of some patterns in Greenberg (1978: 75).

11 Conclusion

In this paper, I have proposed a Distributed Morphology analysis of double case exponence in certain wh-words and wh-based indefinites in Digor Ossetic. This analysis supports the intuition that Multiple Exponence is not a single phenomenon, but can be brought about by a variety of morphosyntactic mechanisms.

Admittedly, the specific phenomenon addressed here is very rare. However, careful examination of typological rara has played a significant role in the development of linguistic theory. By its existence, a rarum reveals a grammatical mechanism that usually remains inoperative or hidden.

The phenomena that are brought to light by the Ossetic double case marking are, first, the possibility for a single nominal to involve several D heads, and, second, a mechanism of feature sharing between such heads.

However, the question always remains of why a given phenomenon is rare in the first place. It follows from the analysis proposed here that to observe this type of multiple case marking, several independent conditions must hold: the presence of two D heads within one nominal, the necessity of feature sharing between them, and lack of dissimilation processes radical enough to fully obliterate one of the case markers. The likelihood that all these conditions will be simultaneously fulfilled is not particularly high.

Competing interests

The author has no competing interests to declare.