In this section, I will present data from various language mixing pairs, see also Hakimov (2021) for further patterns.4 (10–11) show examples from Romanian English and Romanian Spanish mixing pairs. In (10), the English plural noun combines with Romanian plural morphology, and the same is shown in (11), where the Spanish plural noun combines with Romanian plural morphology.

In (12), from Gutiérrez-Bravo & Uth (2020: 91), a Spanish plural noun combines with the Yucatec Maya plural. What is interesting here is that, unlike the patterns involving English and Romanian, plural morphology is optional on Yucatec nouns. A second interesting observation Gutiérrez-Bravo & Uth (2020) make is that the affixes must appear in the order shown in (12); the reverse order is un-attested in their data, and it is also not reported elsewhere in the literature:

A very interesting example of Greek Turkish double plurals is discussed in Joseph (2016), see (13):

The noun xodza-lár-es has both the Turkish -lar- plural as well as the Greek affix -es, which signals both plural and nominative.

Further plural doubling mixing pairs are illustrated below: (1) repeated as (14) from Ewe English, (15) from Hiaki Spanish, (17) from Quechua Spanish, (18–19) from Bantu (Shona and Zulu) English, and (19) from Bantu (Lingála) French mixing:

Thus, double plurals exist in a variety of language mixing pairs.5 In the next section, I proceed to the introduction of my assumptions about the structure of the DP and the basics of Distributed Morphology, and then to an analysis of the Ewe English, Hiaki Spanish and Bantu English/French, and Greek Turkish pairs. I will also discuss how my proposal extends to the Romanian/English Spanish cases. The focus on these pairs is determined by the fact that the sources cited provide descriptions that allow me to offer a morpho-syntactic analysis of the data. Before doing that, however, two issues should be addressed:6 first, the patterns we have just seen are to a certain extent asymmetrical as for as the obligatoriness and the form of plural marking is concerned. While plural marking is obligatory in e.g., English, Hiaki and Romanian, the status of Bantu noun classes as markers of plurality has been controversial in the literature, see Mufwene (1980), Kramer (2015), and more recently Fuchs & van der Wal (2023) to mention a few sources and references therein. Moreover, in languages such as Yucatec Maya, Turkish and Quechua plural marking is not obligatory. While in e.g., English the plural is a suffix, in languages such as Ewe it is a clitic to the preceding word. Since number marking is not a unified phenomenon cross-linguistically, see Wiltschko (2021) for recent discussion, it is not surprising that this non-uniformity transfers to the mixing pairs. The important point is that double plurality appears independently of the form and the obligatoriness of the plural morpheme. This situation seems to contradict Boumans (1998: 91), who claimed that “[t]he likelihood of double marking appears to increase when each language marks the same feature in a different manner, for instance, by means of prefixes and suffixes.”

Second, with respect to the bilingual populations discussed here, the following remarks are in order: according to Bancu (2013), the Romanian English data come from bilinguals in the US, while the Romanian Spanish data come from Romanian immigrants to Spain, and Romanian is classified as the dominant language in both cases. West Africa is generally recognized as an area where multilingualism is the norm and speakers are generally fluent often in more than two languages (see Amuzu & Singler 2014 for an overview), and similar descriptions are given for Bantu speakers (e.g., Myers Scotton 1993). Southern Albania speakers are described as being bilingual and fluent in both languages by Joseph (2016). All Hiaki speakers speak Spanish next to Hiaki, and according to Estrada-Fernández & Guerrero (2007), Spanish is the dominant language. These could be seen as differences that may correlate with the frequency of use of double plurality and the situations it is used in (e.g., family vs. formal contexts, see Bokamba 2012), not its actual production. But we know by now that such gradience is not only characteristic of bilingual populations, it is found among monolinguals as well (see Francis 2022 for recent discussion). For the purposes of this paper, and following the Null Theory approach to mixing, I assume that we do not need a different theory to account for the competence of Ln speakers, irrespectively of the type of bilingual population these may belong to, building on Lohndal & Putnam (2024) and Mahootian (1993). I further assume that in the case of bilinguals, none of the two languages is ever completely turned off, see also Libben & Swieter (2019) for recent discussion and experimental evidence.

For the purposes of my discussion, I assume, as is standard in Distributed Morphology, that all words are complex, they are built based on acategorial roots and functional morphemes/categorizers. The first categorizing head in the process of word formation is responsible for negotiating the meaning roots may receive, Arad (2003), Embick (2010), Marantz (2013), n in our case. Building on Borer (2005), the functional spine of noun phrases includes (at least) the layers illustrated in (20): D is the layer that introduces definiteness, #P introduces the counting function. Plural morphology is in DivP (acting as a divider), but may also surface on n, i.e., plurality is split within the DP, as illustrated in (20), see also Acquaviva (2008), Lowenstamm (2008), Wiltschko (2008, 2021), Alexiadou (2011a, 2021), Kramer (2016), Adamson (2017), Punske & Jackson (2017), Dali & Mathieu (2021) and others, cf. Booij (1996), Ojeda (2005), Collins & Kayne (2023). The two plurals have distinct functions and semantic contribution:

Let me briefly summarize some evidence for split plurality: Acquaviva (2008) discusses plurals in Breton as showing on the one hand properties of inflectional morphology and on the other properties of word formation. Consider (6), repeated below:

Citing Stump (1989), Acquaviva (2008: 239) notes that -e is an obsolete and unproductive stem extension pattern, and then argues in detail that Breton uses for word formation nominal plurality, hence the emergence of double plurals. Robinson (2022) further elaborates on this analysis; the plural morpheme with the greater number of selectional restrictions consistently appears closer to the root in all Breton nominals. -ed, which is lexically restricted to semantically animate nouns, is always inserted closer to the root than -où in a double plural, (21). Such lexical restrictions are accounted for by the proposal that these plurals realize features on n:

In Amharic, see (7) repeated below, double plurals involve a combination of irregular and regular plural forms.

According to Kramer (2016), every noun can have both an irregular and a regular plural, suggesting that the two do not compete for insertion in Number. This fact as well as the presence of double plurals provide support for a split analysis of plurality. From this perspective, while regular plurals realize Number (our Div), irregular plurals realize n, a head that categorizes and specifically nominalizes acategorial roots. As the two plural affixes are associated with distinct heads, they may co-occur, see (5–20). Kramer’s (2016) take on double plurality relies on the fact that regular and irregular forms are not in competition. As she points out, double plurals in Amharic are of the form [Root- irregular plural-regular plural]. Only a subset of nouns has an irregular plural – the irregular plural is not productive across nouns; however, all nouns have a regular plural. Importantly, irregular plurals have selectional restrictions on stems and nouns have an idiosyncratic interpretation when irregularly pluralized. There properties are reminiscent of root-based word formation processes as detailed in Arad (2003), Embick (2010) and Marantz (2013) among others. This leads then to an analysis of Amharic, according to which the regular plural realizes Div7 and the various irregular forms realize n.8,9

As in Punske & Jackson (2017), I further assume that (20) is universal, contra Kramer (2016).10 When it comes semantic contribution of these two plural heads, beginning the lexical plurals, i.e., plurals on n, as discussed in Acquaviva (2008), this type of plurality covers a wide range of interpretations, it is idiomatic and not productive; it creates nouns out of roots (e.g., good-s ‘merchandise’), and it also hosts collectives. Lexical plurality involves a plural from that is associated with n in (20), it creates nouns out of roots. By contrast, grammatical plurality in Div is compositional and productive; it creates pluralities of nouns, and as we have just seen, it appears outside lexical plural in languages that lexicalize both (e.g., Amharic, Kramer (2016)).

According to Acquaviva (2008: 1–2), lexical plurals involve special morphological and semantic properties, which are summarized here and see Alexiadou (2021) for an overview: a) listedness, b) idiosyncratic ending, c) lack of inflectional disjunctivity, i.e., regular plurals are not blocked by the lexical variant, d) in languages that have grammatical gender, lexical plurals go hand in hand with a fixed gender, e) constitute a class of lexically related lexemes, f) lexical plurals are the result of a lexeme forming operation. We will be reminded of these properties in what follows. For Acquaviva, collectives and other minor numbers such as duals and paucals are also lexical numbers: such numbers differ from plurals on Div as far as their reference domain is concerned.11

With respect to Div, its function has been described in Borer (2005) as follows: plurals divide undivided mass, which can then be counted, i.e., embedded under the Quantity Phrase. For Borer, plural in a number marking language such as English has the same contribution as classifiers in languages such as Chinese and does not create individuals. Support for this comes from Armenian, where plural morphology and classifiers are in complementary distribution, as shown in (22), from Borer (2005: 94–95):

Acquaviva (2008) offers a detailed discussion about meaning differences between the two heads in (20), see also Punske & Jackson (2017) and Wiltschko (2021). Specifically, Acquaviva (2008: 270) states: “When number is entirely a property of the syntactic context, it is encoded as the head of the DivP, directly above the noun-denoting complex [nP n [ Root ] ]. When a particular choice of number is instead inherent in a noun, the relevant information is expressed on [n] itself.”12

Thus, if indeed there are two loci for plural morphology with distinct contribution and function, namely DivP and nP, then double plurality can be captured straightforwardly, while its treatment is problematic under an approach, where there is only one NumberP, Ritter (1992). The only way to accommodate double marking by assuming a single locus is to employ Fission, a post-syntactic operation that splits nodes into two. This happens when a particular Vocabulary Item does not discharge all the node’s features (Noyer 1997, Halle & Marantz 1993). As argued in Kramer (2009), Fission cannot be employed at least in Amharic, as it cannot copy a feature into a new terminal node, there would be only one feature [+PL]. Matters may differ if plurality is decomposed as in Harbour (2014), where plural is associated with the feature [–minimal, –atomic], see also Martí (2020). However, since a) there are certain restrictions on double plurality, which would be unexpected if Fission were employed, and b) the one plural form is not interpreted as plural in this sense, a Fission account will not be entertained here.

As double plurality does not always occur in Amharic, Kramer argues that plural n and plural Div may optionally undergo Fusion, defined as a process merging two syntactic terminals into one. This can be viewed as following from the principle in (23) (Siddiqi 2006: 82). According to Siddiqi, Minimize Exponence forces Fusion to occur; Fusion is motivated by an effort to reduce an utterance to the fewest pronounced morphemes, and is in contrast with the bi-uniqueness principle in (7):

What I would like to argue is that double plurals in mixing provide further support from the split analysis of plurality. Specifically, building on Martin et al. (2023), Guasti, Alexiadou & Sauerland (2023), cf. Hein et al. (to appear), I propose that double plurals adhere to (8), repeated below:

From this perspective, there is no such thing as multiple exponence or redundant exponents in the context of double plurals: double marking involves realization of plurality in two distinct heads, n and Div, and the two heads receive a distinct realization each. In fact, double plurals are expected under such a decomposition. Thus, while fusion eliminates multiple exponence, the reverse is observed in the contexts we are interested in, providing cases of multiple exponence where the two exponents are not redundant.

In what follows, I will analyze double marking in Ewe English, Hiaki Spanish and Bantu English/French and Greek Turkish mixing as instances of split plurality. This is in line with recent work that has successfully applied the Distributed Morphology perspective on word formation to language mixing e.g., Alexiadou (2011b), Alexiadou et al. (2015), Alexiadou & Lohndal (2018; to appear), López (2020; to appear), Lohndal & Putnam (2021, 2024) and many others. There are no additional operations that form words in multilingual grammars. The main feature of multilingual speakers is that they may realize features with exponents drawn from both their lexical inventories. Specifically, following López (2020), I assume that the lexicon of bilingual speakers is integrated: they have one List 1, containing roots and abstract features, and one List 2, containing the Vocabulary Item rules (VIRs) at their disposal from both their languages, see also Alexiadou & Lohndal (to appear). As these authors argue, List 1 items are employed to build words and phrases. List 2 provides exponence to features. Following Embick (2015), and Alexiadou & Lohndal‘s (2018) as well as López‘s (2020) implementation of mixing in Distributed Morphology, each terminal is associated with semantic features, drawn from a universal feature inventory. Prior to VIR, Fusion may take place. When VIRs apply to a syntactic node, the exponent occurs in the position of the node. The properties of Vocabulary Insertion are illustrated in (24):

As an illustration, consider tense formation in English: the competition for insertion at the T node is negotiated between the three VIs in (25): in any given derivation, only one may be employed and inserted in the structure in (26).

In certain cases, e.g., broke, an additional Readjustment rule is required. Such rules are defined as morpho-syntactically triggered phonological rules that alter the phonology of roots.

When it comes to providing exponence to terminals, Alexiadou & Lohndal (2018) and López (2020; to appear) argue that bilingual speakers simply have more options: they could provide two realizations for roots and abstract morphemes, one from each language, and also combine roots with functional heads bearing different flavors.

In this section, I will substantiate the hypothesis that one of the plural affixes is used to create a word out of a root. According to Amuzu (2009: 153ff.), the following restrictions apply when it comes to double plurals in Ewe English mixing:

Amuzu concludes that Ewe wó is the only plural marker. Amuzu (2013) presents similar restrictions involving Ewe French language mixing. The observation that the -s plural appears only on English words is reminiscent of the discussion in Fisher et al. (2022): -s in Pennsylvania Dutch also occurs predominantly with English roots. In the examples Amuzu discusses, the point is that -s appears to alternate with zero realization. Assuming Maximize Exponence, the one head that can be lexicalized by -s or zero is n. Especially in view of (27d) and (27e), Amuzu’s insights can be straightforwardly accommodated under split plurality: (27d) shows that irregular plurals combine with regular plurals, and (27e) that collective nouns, which carry the -s affix may not be pluralized by a grammatical plural. In Ewe, collective and mass nouns do not combine with wó, e.g., tsi ‘water’.13 When they do, the plural form receives the different kind interpretation provided by grammatical plural (e.g., kinds of water, Dzablu-Kumah 2015). Amuzu makes the following observations (Amuzu 2013: 87): “An English noun that occurs in this kind of structure is one whose referent naturally occurs in pairs or groups, and/or when… [it] is generally referred to collectively. When such nouns, however, take both -s and -wó, they would signal that different kinds or different groups of the entities in question are being referred to. For example, SANDAL-S-wó will mean ‘the pairs of sandals’”. In this respect, the English plural collectives behave similar to the Ewe bare collectives. Collectives are lexical plurals in Acquaviva’s (2008) analysis, i.e., they cannot associate with plurality in Div. The interpretation of sandal-s, the form containing the lexical plural, in combination with wó, the grammatical plural, aligns with the semantic contribution of Div to nP. From the perspective of split plurality then, the English plural is the realization of plural on n, while the Ewe plural is the realization of Div, see (28a) and the VIRs in (29a–b):

To get the correct word order, we must assume nP raising at least to DivP, as detailed in Aboh (2004) for languages such as Yoruba and Gungbe. Specifically, according to Aboh (op.cit.) number morphology is realized under Number (our Div), but as wó appears in final position within the noun phrase, further phrasal movements may be necessary to yield the correct word order.

The Hiaki Spanish situation is very similar. According to Pierson (2017: 48): “many Spanish forms are loaned into Hiaki by semantically bleaching the plural -s on a Spanish word.” Two pieces of evidence suggest that the Spanish plural is not plural in Div, and thus not carrying the semantics of grammatical plurality, but rather a lexicalized affix, a plural on n: First, as shown in (30), Spanish plural nouns can occur with singular definite determiner, suggesting that the plural form does not carry grammatical plural meaning, (30a). The form may be further pluralized with the Hiaki affix -im, and be accompanied with the plural Hiaki definite determiner, (30b):

Borrowing of the Spanish plural applies randomly to certain nouns, see (31), which is reminiscent of the properties of lexical plurals discussed in 3.1.

According to Estrada-Fernández & Guerrero (2007: 421), an explanation for this might be that the Spanish forms are introduced as collectives. As seen above, and discussed in Acquaviva (2008: 257), “the term collectives refers to nouns that are plural by syntactic and morphological criteria, which crucially, however, differ from regular count plurals in the conceptualization of their reference domain.” These are analyzed on a par with minor numbers as lexical plurals, see section 5. If these forms are collectives, then we may view the Spanish plural as a plural on n, the locus of creating a collective interpretation (Acquaviva 2008). The above mixing data may receive the representation in (32a) and the VIRs in (32b–c). N-movement to Div yields the order seen in (31b):

Let me now turn to double plurals in Bantu English/French mixing. As Mashiri (2002) details, the English plural marker co-occurs with Shona plural class marker 6 ma-, (33a–b). According to Mashiri (2002), trying to leave out ma- but retaining -s produces ungrammatical sentences, see (33c–d). Mashiri stresses that ma- is compulsory to complete the morphosyntax:

According to Kamwangamalu (1997), any English plural noun may be prefixed with the class marker 6 ama- in Zulu English mixing. Fuchs & van der Wal (2023: 310–311) note that in Gikuyu English mixing the plural affix appears as expected, but English -s must also be present, see (34). Class 6 in Bantu includes plurals of class 5, and collectives, liquids, mass nouns (see Demuth 2000).

Whenever English nouns are borrowed in Bantu, they must appear with Bantu morpho-syntax, and typically they enter class 5, see (35) from Tappe (2013: 270), and Amuzu (2013):

In Lingála, (36a), double marking is not constant in mixing but depends on interlocutor, context of speech or the occasion (Bokamba 2012: 299). Bokamba (1988) and Kamwangamalu (1997) report that these double plurals had the effect that double plurality is now systematic in the Kinshasa Lingála variety, while this is not the case in other Bantu languages apart from Kikongo. Specifically, in this variety, the plural prefix of the human class ba has spread to all noun classes, the result being that all other classes except classes 1/2 undergo pluralization twice, Bokamba (1988: 55), see (36b):

According to Kihm (2005), Ferrari (2005), Kramer (2015), López (2020), Fuchs & van der Wal (2023) and others, but cf. Déchaine et al. (2014), Taraldsen et al. (2018), Bantu class markers realize n. It is assumed that the singular and plural of the same noun have the same noun class, and the realization of the noun class varies depending on number, as illustrated in (37), from van der Wal & Fuchs (2019):

From the descriptions available to me and adopting the view that class markers are realization of n, then rule (37) applies in the mixing cases as well. The question is what about the English/French plurals? Muysken (2011) states that -s is pied-piped along with e.g., the English word, and it is not crucial for the meaning of plurality, which is realized by the relevant Bantu class marker. If -s functions as a word marker, then it is possible to analyze it as also being on n, as in (38a). It is well-known that some noun classes in Bantu may stack. Bokamba (1988: 56) suggests that double plurals in the Kinshasa Lingála variety are the result of the reanalysis of the foreign affix as a Bantu prefix, followed by attachment of the overgeneralized Bantu marker ba to the derived noun. This insight supports a view of the foreign plurals as word markers (but see the remarks in section 5 on other pairs including French). Two further alternative analyses are: a) (37) applies and English/French plurals spell out PL in Div in addition, as in (38b). Root to n followed by nP to Spec, DivP movement would give the correct word order. b) Fuchs & van der Wal (2023: 310) suggest that one language (English) must be closed off by NumberP (our DivP) before the other language (Gikuyu) can be switched to, alluding to a version of (38a), where it is DivP and not nP that appears under nP realized by ma.

The Greek Turkish example xodza-lár-es discussed in section 2 can also be accommodated under split plurality, building on Markopoulos’s (2019) treatment of imparisyllabic nouns in Greek. According to Joseph (2016), the double plural form co-exists with the form hodz-ad-es, where the plural is only realized once by the Greek suffix. Note, however, that the plural of this noun class cannot be built without -ad-. The -ad- affix is found in many Greek nouns, but only in the plural, creating so called imparisyllabic forms, whereby the number of syllables in the plural is greater than that in the singular. Joseph (2016) argues that it is very well possible that Turkish -lar- has been recruited as having a similar function to – ad -. As Joseph (op.cit.) points out, the Greek affix -es appears either following -ad- or following -lar-. He compares this borrowing with the borrowing of the Turkish past Tense form -dI- into Greek, which is now a simple verbalizer. Joseph concludes that Turkish -lar- is not really a plural marker per se. Markopoulos (2019) argues that when one zooms in on the morpho-syntax and semantics of the Greek affix -ad-, one notes that its distribution is restricted: it only applies to Greek [+human, –feminine] nouns. Building on Kramer (2015), Markopoulos proposes that -ad- is the exponent of an n head with the features [+human, –feminine]. This leads to the representation in (39), where lar and ad are in complementary distribution realizing n and the Greek plural is in Div:

The Romanian/English and Romanian/Spanish pairs may also be treated along the lines of what has been suggested for Ewe English: the English/Spanish forms sales and offers are lexical plurals in Acquaviva’s (2008) sense, as pointed out by an anonymous reviewer, and thus may be further pluralized by the Romanian affix in Div. Finally, a split analysis may apply to double plurals in the case of the English Russian example jeansy, provided by an anonymous reviewer. As in the other cases, there is no evidence that -s in this form is a plural marker. Indeed, jeans is a lexical plural in Acquaviva’s (2008) sense, in which case the root jean and -s combine to form the noun jeans, which can then be affixed by the Russian plural marker.14