The basic plural suffix in Barguzin Buryat is -nuud, which I gloss as PL1. This plural marker can appear in nominals of any case—nominative, accusative, genitive, as well as the various obliques. Since the distribution of this plural suffix has no restrictions, I refer to it as the ‘basic’ plural form. In (5) below we see this morpheme previewed in accusative3 and genitive contexts:

The basic plural -nuud contrasts with its more restricted optional variant -nuuʃA, which I gloss as PL2 to distinguish it from the basic plural. The capital ‘A’ in -nuuʃA represents a harmonizing low vowel. As I discuss in section 3.1 below, Barguzin Buryat has vowel harmony, and this harmonizing vowel /A/ appears in many morphemes. Speakers describe -nuuʃA as a dialectical or colloquial suffix specific to their regional variety of Buryat. There is no motivation for a phonological explanation for the alternation between -nuud and -nuuʃA (as I discuss in detail in section 4.1 below), nor is there semantic difference between these two plural forms. Therefore I regard -nuuʃA as a contextually-triggered suppletive expression of the plural. Importantly for this paper, while -nuud can appear in any context, the -nuuʃA plural is limited to accusative and genitive contexts. These grammatical uses of -nuuʃA are previewed in (6) below:

The impossibility of -nuuʃA in nominative contexts is demonstrated in (7a) below. This restriction is not surprising for the theories about case and suppletion summarized above. As mentioned previously, what is puzzling for the relevant theories is the further fact that -nuuʃA also cannot occur with oblique cases, as (7b) below shows in a dative context:

Since -nuuʃA can occur in accusative and genitive contexts, its impossibility in oblique contexts violates the prediction in (3) above. This is the challenge that this paper is concerned with.

Notice that both the -nuud and -nuuʃA plurals contain a sub-part -nuu. As I show in section 4.2, there is evidence that this is an independent element, since it can be excluded from some plural forms. Therefore I will factor this morpheme out in the final analysis, which I will frame in terms of an alternation between two plural variants -d and -ʃA. For simplicity of exposition, I will speak in terms of -nuud and -nuuʃA for the first part of the paper, and justify their decomposition later on. Setting this detail aside for now, the facts that this paper is concerned with are summarized in (8) below. As the paper progresses, we will deal with a number of additional intricacies about Barguzin Buryat plural marking, but (8) accurately describes the core patterns.

The majority of recent literature on the *ABA generalization and related topics uses one of two frameworks—Distributed Morphology, or Nanosyntax (Starke 2009; Caha 2009, a.o.). Though I will discuss a Nanosyntactic approach to these Buryat facts at the end of the paper, I will focus on an implementation using Distributed Morphology. For this approach, the syntactic derivation builds an abstract hierarchical structure and then passes it on to the PF component of the grammar. This component then assigns morpho-phonological form to the terminal nodes of the syntactic tree depending on the features they bear, by referencing a listed set of Vocabulary Insertion (VI) rules.

In classic Distributed Morphology, the process of assigning morpho-phonological form proceeds terminal-by-terminal, and thus in the basic case one morpheme cannot correspond to more than one terminal node. However, there are indeed situations in human language where a single morpheme seems to express the features of multiple terminals. Such morphemes are sometimes termed portmanteau morphemes, and these will play a central role in this paper.

To achieve portmanteau formation, much literature using Distributed Morphology appeals to a mechanism of fusion, which unites multiple terminal nodes into one before morpho-phonological assignment occurs. As previous research has noted, fusion has the problematic property of requiring the grammar to know which terminal nodes to fuse prior to the application of the relevant VI rule—in other words, a ‘look-ahead problem’ (Chung 2007a, b; Caha 2009, 2018). For this reason, here I will eschew fusion. Instead, I will implement portmanteau formation by spanning, which allows a VI rule to target multiple terminal nodes that form a contiguous sequence (Bye & Svenonius 2012; Merchant 2015; Haugen & Siddiqi 2016; Svenonius 2016, a.o.). This allows a single morpheme to sometimes simultaneously express the features of multiple terminals, as needed. We will see spanning in action in the analysis of section 5, which I preview next.

As discussed above, if the generalization about suppletion in containment hierarchies (1) and the case containment hypothesis (2) are both correct, we expect the consequence in (3), repeated in (9):

If we find a pattern that violates (9), there are two main possibilities. On one hand, (1) or (2) might simply be incorrect. On the other hand, it is possible that (1) and (2) are correct, but that independent factors can sometimes prevent them from interacting in the usual way. I argue that this second type of analysis is accurate for plural suppletion in Barguzin Buryat.

I argue that -nuuʃA is a portmanteau, whose feature specification overlaps with that of oblique morphology. This prevents the two from co-occurring, thus yielding a superficial ABA pattern. The portmanteau-hood of -nuuʃA is revealed by its interaction with accusative / genitive case morphology. Notice that in (5) above, accusative and genitive morphology (here respectively -iijɘ and -ain) affix straightforwardly to the basic plural -nuud. However, in (6) above, the suppletive plural -nuuʃA appears without the typical accusative or genitive marking that we saw in (5). In fact, combining -nuuʃA with typical accusative or genitive morphology is unacceptable, as (10) shows.4

I hypothesize that -nuuʃA blocks accusative / genitive case affixes because -nuuʃA is a portmanteau of plural features, and accusative / genitive features. Assuming that a given syntactic feature can only be morphologically expressed once (Bobaljik 2000), since -nuuʃA alone expresses all of these features, independent accusative / genitive marking need not, and cannot, occur with it.

With this hypothesis in mind, notice that according to a case hierarchy like that in (2) above, oblique cases involve a syntactic structure including nominative as well as accusative / genitive features. Correspondingly, in this paper I will posit that oblique suffixes in Barguzin Buryat are portmanteau morphemes that express all of these case features. Importantly, if this is so, then -nuuʃA and oblique morphology overlap in their feature specifications: both must express accusative / genitive features. I argue that for this reason, -nuuʃA and oblique morphology cannot co occur: since each must express features that the other also depends on, they have a complementary distribution.

For an initial illustration of this proposal, see (11) and (12) below. In (11), we see the structure for a plural oblique nominal as analyzed under the simplified case hierarchy I will adopt here, which I explain in section 2. In (12), we see a preliminary set of relevant VI rules for Barguzin Buryat. Both (11) and (12) will be slightly modified later on, but these will suffice to make the main point clear. (Note that accusative and genitive cases have been bundled together in (11-12). I justify this decision in section 2 below.)

Several of the VI rules in (12) above describe morphemes that correspond to multiple adjacent terminals—a possibility allowed by the spanning hypothesis, as mentioned above. Importantly, since the rule for -nuuʃA (12b) and oblique morphology (12e) overlap, I argue that both cannot apply in the same nominal domain. This morphological conflict prevents them from co-occurring, and yields the superficial ABA distribution of -nuuʃA. In contrast, notice that the VI rule for the basic plural -nuud (12a) and oblique morphology (12e) do not overlap. Therefore both can be inserted into an oblique nominal structure, as shown once again in (13):

Though we have identified a reason why -nuuʃA and oblique morphology cannot co-occur, inserting -nuuʃA alone in (11) would successfully express almost every feature in the functional spine of the nominal, aside from [OBL]. In this situation, the rule for oblique morphology (12e) could not apply, and we would expect to end up with an oblique nominal containing -nuuʃA where oblique morphology fails to occur. In reality, such forms are unacceptable (14):

Notice that in (13), the combination of -nuud and oblique morphology expresses all features in the functional spine of the nominal, while use of -nuuʃA alone as in (14) leaves an [OBL] feature unexpressed. I argue that for this reason, forms like (13) are always selected over those like (14) because of a blocking effect (Aronoff 1976; Embick & Marantz 2008; Gardani et al. 2019, a.o.). Specifically, I adopt the view of morphological competition in Middleton (2020), who uses spanning to analyze patterns in pronominal syncretism. Middleton argues that the combination of morphemes assigned by VI (within a given syntactic cycle) is the one that most completely expresses the structure in question. This general idea also has precedent in Haugen & Siddiqi (2016). As I discuss in section 5, this theory makes exactly the right predictions about Barguzin Buryat.

In summary, a variety of independent factors prevent -nuuʃA from ever occurring in oblique nominal environments. Thus -nuuʃA has an ABA distribution. However, this pattern is fundamentally an epiphenomenon which does not falsify the theories that ban ABA patterns under normal circumstances. Rather, it reveals a way that ABA can exceptionally arise even in the context of such theories, as the rest of this paper argues in detail.

Next, section 2 provides background on the *ABA generalization and theories of case containment. Section 3 describes the relevant facts about Barguzin Buryat morpho-phonology. Section 4 describes the plural morphology of this language in detail, and shows why the -nuud/-nuuʃA alternation is not phonological. Section 5 provides the main analysis using Distributed Morphology. In section 6 I also discuss a Nanosyntactic analysis. Section 7 contains the concluding remarks.

Analyzing the morphology of Barguzin Buryat requires familiarity with a few phonological processes, reported here following the description in Staroverov & Zelensky (To appear). This paper adopts the transliteration system used in that work (as well as in Tatevosov et al. To appear), which is an IPA-based representation of the original Cyrillic Buryat orthography. In careful speech the diphthongs transliterated as ⟨ei⟩, ⟨əi⟩, ⟨oi⟩ and ⟨ai⟩ are pronounced as expected following the IPA, but in more natural colloquial speech, the first three diphthongs are simplified to [e:], and the latter to [ɛ:]. This language also has vowel harmony, but the details of this process do not affect the morphological facts under examination here in any significant way. It is only necessary to be aware of the harmonizing low vowel /A/, which is realized as /a/, /ə/, or /o/, depending on the phonological properties of the stem that it affixes to.7

The forms created by agglutinating nominal morphology in this language are frequently affected by its two strategies for avoiding hiatus (vowel-vowel sequences). First, when a heavy vocalic segment (long vowel or diphthong, consisting of more than one mora [=<μ>]) is adjacent to a short vowel, the short vowel deletes, as shown in (18):

Second, when two heavy vocalic segments are adjacent, neither is deleted. Rather, the segment /g/ (phonetically often [ɣ/ʁ]) appears between them, as (19) exemplifies. This is a typologically unusual epenthesis strategy, which is subject to some qualifications as Staroverov (2016) argues, but the level of description in (19) is sufficient for this paper.

As is cross-linguistically frequent, nominative case in Barguzin Buryat is null. Oblique cases involve straightforward suffixation of -tA (dative), -tAi (comitative), -AAr (instrumental), or -aan/-χAA (ablative, which has two free variants). We will see these suffixes in many following examples.

In contrast, accusative and genitive marking are more complex, in a way that is phonologically determined. When affixing to a nominal form ending in a long vowel or diphthong, accusative case is -jə (20a-b), while genitive case is -n (20c-d):

However, when suffixing to a nominal form ending in a short vowel or consonant, accusative case marking is -Aijə/-iijə, while genitive case marking is -Ain/-iin, as we see in (21) below. Since these accusative and genitive forms have an initial heavy vocalic segment, when affixing to a nominal form ending in a short vowel the hiatus avoidance process in (18) above deletes that short vowel, as (21c-d) below show.8

It is descriptively correct to hypothesize the following: Fundamentally accusative marking is -jɘ, and genitive marking is -n. Both of these morphemes must affix to a heavy vocalic segment. When the nominal form being affixed to does not end in a heavy vocalic segment, an epenthetic element Ai/ii is inserted to satisfy this need. It is well-known that morphology can be sensitive to phonological context in ways such as this (McCarthy & Prince 1993, 1998; Arregi & Nevins 2012), though alternative analyses of these Buryat facts are conceivable. What I have said here is sufficient for the purposes of this paper, however.

With the relevant morpho-phonological background now laid out, we are prepared to examine in detail the patterns of plural marking that this paper will analyze.

The alternation between -nuud and -nuuʃA is not the result of a phonological process. The most straightforward phonological analysis of these plural forms would be to consider -nuuʃA a form derived from the plural -nuud plus a special case morpheme -ʃA that is syncretic for accusative and genitive, whose presence triggers deletion of the final /d/ of -nuud due to a phonological process that simplifies the potential [dʃ] cluster. Consistent with such an analysis is the fact that -nuudʃA is not a possible accusative or genitive form, as (28) shows:

However, clusters with a consonant + [ʃ] are generally permitted in Barguzin Buryat, and indeed, forms with [dʃ] are possible outside of contexts like (28). This can be seen by combining the 2nd person singular possessive marker -ʃni with various nominal forms ending in /d/, as in the examples of (29). Most important of these is (29a), where we see the plural -nuud combining with such possessive morphology without any deletion:12

Since [dʃ] is permitted by the phonology of this language, there is no obvious phonological explanation for the -nuud/-nuuʃA alternation. Thus I take this alternation to be morpho-syntactically conditioned suppletion. Given this conclusion, this alternation stands as a puzzle for the theories of case containment and suppletion described earlier in this paper.13 Before proceeding to the analysis, next I will consider the morphological decomposition of -nuud and -nuuʃA in more detail.

Here I will discuss a final fact about the morphological structure of plural marking in this language. So far, I have spoken in terms of two plural forms -nuud and -nuuʃA. These both contain a sub-part -nuu. In principle, it is conceivable either that this is a synchronic coincidence (perhaps with a diachronic explanation), or that -nuu is in fact a separate morpheme in the synchronic grammar. There is evidence suggesting that the latter hypothesis is the correct one. In particular, with certain nouns (often animate ones, but not only) -nuu may be excluded, leaving behind -d as the only expression of plurality, as in (30):

This suggests that -nuu is a separate morpheme, and that the basic plural marker in Barguzin Buryat is in fact -d. If this is so, then when we factor out -nuu, we come to the conclusion that the -nuud / -nuuʃA alternation is more fundamentally an alternation between two elements -d and -ʃA. This hypothesis accurately predicts the fact that the short plural -d can alternate with an alternative short plural form -ʃA, as demonstrated in (31) below. In (31a-b) we see nouns using the short plural -d in a nominative context, while in (31c-d), we see the same nouns in accusative contexts using a short plural -ʃA.

Thus we have evidence that -nuud and -nuuʃA actually contain a separate element -nuu that is correlated with plurality but not necessarily a plural marker itself.14 Consequently, I propose that the actual plural suffixes in Barguzin Buryat are -d and -ʃA. I will thus factor -nuu out of the coming analysis. This decision does not alter the puzzle that this paper focuses on. Given the relevant theories described earlier in the paper, any morphological process that is available in accusative and genitive cases, but not oblique ones, is unexpected. Since -nuu was present in all the plural examples reported in this paper until now, the puzzle that those facts pose is not affected by uniformly factoring out -nuu. Once this is done, the puzzle is conceptually the same, though cast in terms of -d versus -ʃA rather than -nuud versus -nuuʃA.

There are multiple ways of analyzing -nuu. For instance, this element could be part of a decomposed structure for number (Harbour 2014, a.o.) or an allomorph of a functional head like n⁰ (Embick & Marantz 2008; Embick 2010, a.o.) in plural contexts (though able to be omitted for certain nouns, as we’ve seen). However, the nature of this element does not have a direct bearing on the examination of the -d / -ʃA alternation. Thus in the coming analysis, for simplicity of exposition I will opt to diagram -nuu as a sub-part of the noun, beneath the nodes encoding number and case. With this final consideration addressed, we can now summarize the full set of relevant Barguzin Buryat facts as follows (32):

Given the case containment hypothesis, the fact that the -ʃA plural cannot occur in nominative contexts is expected, since nominative structures do not contain features related to accusative / genitive cases. However, accusative / genitive features are hypothesized to be a sub-part of oblique cases, which is why the impossibility of -ʃA in oblique contexts is surprising. In the next section, I will make explicit the analysis previewed in the introduction—that a morphological conflict is responsible for the fact that oblique morphology, and the -ʃA plural, have a complementary distribution. As a result of this conflict, only the basic plural -d is ever seen to co-occur with oblique morphology, and -ʃA thus has an ABA distribution.

Defining the VI rules necessary for this analysis will be facilitated by addressing a question for theories of case containment: if a case containment hierarchy like the one I have adopted in this paper holds cross-linguistically, then we must ask why the structural complexity attributed to many cases under such a theory is rarely surface-evident in a straightforward way. Smith et al. (2019) show that case morphology is sometimes surface-evidently complex in the expected way for two languages, Khanty and Kalderaš Romani (33). Nevertheless, the fact is that in most languages with overt case morphology, each case is simply expressed by one morpheme. For this reason, much recent work on case containment relies on relatively indirect evidence from phenomena like suppletion and syncretism.

I hypothesize that for languages like Barguzin Buryat with mono-morphemic case marking rather than surface-evident containment, all features of the case hierarchy present in a given nominal structure are expressed by a single portmanteau morpheme.15 This is essentially the view taken in Caha (2009; 2013), whose Nanosyntactic approach to case entails that most case morphemes are mapped to a constituent containing several nodes of the hierarchy. As mentioned in the introduction, this paper will set Nanosyntax aside until section 6 below, instead focusing on a Distributed Morphology account in which portmanteau morphemes are formed by spanning—a mechanism that allows a single VI to ‘stretch’ across multiple contiguous terminal nodes.

Given these proposals, and following the version of the case hierarchy justified in section 2 above, we can state the VI rules for case morphology in Barguzin Buryat as in (34a-c) below. These rules state that nominative case expresses the feature [UNM] (34a), accusative and genitive case express the feature set [UNM DEP] (34b), and oblique cases express the set [UNM DEP OBL] (34c). In (34d-e), we also see the VI rules I posit for plural morphology. As previewed in the introduction (though now factoring out -nuu), I argue that the basic plural -d simply expresses a plural number node (34d), while the -ʃA plural is a portmanteau, as we see in (34e). Specifically, I argue that -ʃA expresses both a plural feature as well as the features corresponding to accusative / genitive case, which in the context of this account, are the features [UNM DEP]. This proposal accounts for the fact that the -ʃA plural bleeds the appearance of independent accusative / genitive case morphology, but can occur in contexts where those cases are typically assigned, provided that their corresponding morphology is omitted. We have seen this, for instance, in (23-24) versus (26) above.

We are now prepared to explain the patterns of Barguzin Buryat plural nominal morphology described in the previous section.

Here I will discuss the derivation of each plural form one by one, which will lead straightforwardly into my explanation for why the distribution of -ʃA is restricted. For concreteness, I will assume that after a given syntactic structure is built and passed on to the PF component of the grammar, its terminals are then assigned linear order, after which VI rules apply (Embick 2010; Arregi & Nevins 2012; Haugen & Siddiqi 2016; Ostrove 2018; Davis 2020).

For a plural nominative nominal, the structure in (35a) is built. When completed and evaluated by the morpho-phonological component of the grammar, the linearization of that structure and the application of VI rules to it yields the representation in (35b). Here the plural number node is realized by -d, and the lone case node bearing [UNM] is assigned -∅, consistent with the fact that nominative case in Barguzin Buryat is systematically null. Note that here and throughout this section I ignore the realization of N, since this does not interact with the plural facts in focus here.

In this context, there is no possibility of using the -ʃA plural. As the VI rules in (34) above state, the features upon which the use of this morpheme depends are not present here. However, use of -ʃA becomes a relevant possibility when we consider accusative / genitive nominals.

Following the version of the case hieararchy assumed in this paper, a plural accusative or genitive nominal has the same structure as a nominative one, aside from the addition of the next node up in the case hierarchy, which bears the feature [DEP]. This is shown in (36a) below. In (36b), we see the morpho-phonological form that is assigned to this structure in situations where the basic plural -d is used. In this context, following the VI rules defined in (34) above, -d expresses the plural node, while accusative / genitive morphology is inserted as a span that subsumes the two case nodes [UNM] and [DEP]. The facts have shown us that there is also another way to morpho-phonologically express the nominal structure in (36a). As the VI rules in (34) above state, such a structure can also be expressed by the -ʃA plural portmanteau. This is a span that realizes all of [#_pl], [UNM] and [DEP], as (36c) shows:

Both of these strategies for realizing such a nominal structure are grammatical in Barguzin Buryat. Let’s consider this fact more deeply, since it connects to concepts that are vital for my explanation of the important puzzle that the -ʃA plural cannot arise in oblique contexts.

Human languages sometimes have multiple ways of realizing a given cell in a morphological paradigm—a state of affairs that Thornton (2011; 2012) terms overabundance. We can understand this fact as a consequence of two axioms of the Distributed Morphology framework: the elsewhere principle (Kiparsky 1973) and the subset principle (Halle & Marantz 1993, a.o.). These principles work together to mandate that the morpheme that is chosen to realize a given terminal node is the one that matches the largest subset of that node’s features. Importantly, it is possible for multiple morphemes to happen to correspond to equally large subsets of the features that a given terminal has. In this situation, both morphemes would be grammatical choices for expressing that terminal (as posited by Hein 2008; Halpert 2016; Driemel 2018). This would give rise to an instance of overabundance. A straightforward situation of this sort is the Barguzin Buryat ablative case, which as described above has two free variants: -aan and -χAA.

This understanding of morphological optionality is not directly applicable to the Barguzin Buryat plural forms in (36b-c) above: here we do not see multiple ways of realizing one terminal, but rather multiple ways of realizing the entire functional spine of the nominal. Importantly, Middleton (2020) extends the principles of Distributed Morphology to situations of precisely this sort in her spanning analysis of pronominal syncretism. Specifically, Middleton (p. 59, 68) argues that within a single spell-out domain, such as the functional extend projection of the noun, the combination of morphemes (some of which may be spans) is chosen that most thoroughly expresses the structure in question.17 Haugen & Siddiqi (2016) make a similar proposal in their analysis of the nature of spanning. On these grounds, the two Barguzin Buryat forms in (36b-c) are equally good choices, since both completely express the functional spine of the nominal. It is unsurprising, given these considerations, that both options are acceptable.18 This theory of morphological competition also explains the impossibility of -ʃA in oblique contexts, as I describe next.

A plural oblique nominal will have a structure like that in (36a) above, but with the addition of the [OBL] feature, yielding a fully articulated case layer. We see this below in (37a). We have seen that this structure can only be realized in one way: with the basic plural -d and usual oblique morphology, the latter of which relizes all features present in the case hierarchy, as (37b) shows:

We saw in detail in the introduction and in section 4 (ex. 27) above that two other conceivable ways of realizing this structure are impossible. Next I provide an explanation for these facts.

First, it is not possible for the -ʃA plural to co-occur with oblique morphology, as demonstrated once more in a dative context in (38):

Assuming that a given syntactic feature can only be morphologically expressed once (Bobaljik 2000), I argue that -ʃA and oblique morphology cannot co-occur because they overlap in the features they must express, given the VI rules defined in (34) above. To make this overlap clear, the features these morphemes express are shown in (39) in the context of an oblique nominal structure. Here we see that the features [UNM DEP] are where the overlap occurs:

Second, while -ʃA and oblique morphology cannot co-occur, in principle it should be possible to assign -ʃA in the structure in (37a) above, and then simply not insert oblique morphology. This derivation would avoid the overlap problem. However, we have seen that such forms are not possible. We see this again in (40), an ungrammatical dative context including -ʃA but omitting dative marking:

I propose that morphological competition is responsible for the unacceptability of examples like (40). Following Haugen & Siddiqi (2016) and Middleton (2020), continue to assume that a structure must be expressed by the set of morphemes that most thoroughly realizes it. This hypothesis accurately rules out the form in (40): here the features [PL UNM DEP] are expressed by -ʃA, but a lone feature [OBL] is left unrealized. This contrasts with the form schematized in (37b) above, which expresses every feature in the functional projection of the noun by combining the basic plural -d with usual oblique morphology. Thus forms like (37b) block ones using only -ʃA such as (40).19

This concludes the main analysis. I have argued here that the ABA distribution of -ʃA is caused by its portmanteau-hood, which brings it into conflict with the morphological needs of oblique suffixes, and interacts with more general principles of morphological competition. Since this ABA pattern arises due to a confluence of independent factors, it does not falsify the morpho-syntatic theories discussed earlier in this paper for which ABA patterns are predicted to be unattested. Rather, these findings reveal a way that portmanteau morphemes can interfere with the principles that give rise to the *ABA generalization under normal circumstances.20 See section 7 for further discussion.