2.1 Morphosyntax of PT nouns

All nouns that head an NP must be marked for number or case, as shown for the object noun kuta ‘stick’ in (5a)—in object position, it cannot occur without either a plural marker or the singular accusative case suffix. When a noun is incorporated (5b) or is the left-hand member of a compound (5c), it cannot be marked for number or case (Haugen & Harley 2012).

This pattern holds for PT nouns as well. They must bear the plural number sufix when they head a separate NP (6a, c), and must lose their number inflection when incorporated (6b) or compounded (6d).

As we have seen in (2), in Hiaki, nominal number is marked on the head noun and on the determiner/demonstrative. As we saw above (3), and illustrate again in (7), number marking on countable pluralia tantum nouns in Hiaki is purely formal: Pluralia tantum forms with numeral ‘one’ are fine, (7a), in contrast to the incompatibility of ‘one’ with plural forms of regular count nouns, which have a plural entailment (7b). It is also worth noting that pluralia tantum concord appears on postposed adjectives and numerals (7c), and even on stranded adjectives and numerals in N-ellipsis constructions (7d).

This “one [PT noun]” construction is, so far as we know, acceptable for all pluralia tantum count nouns in Hiaki, including for paired body part nouns such as puusim ‘eye(s)’ and wokim ‘leg(s)’.

2.2 Collecting and categorizing Hiaki PT nouns

In previous documentation, Hiaki PT have not received much attention, although they are often remarked on in passing. It is an old pattern in Hiaki. The earliest extant documentation, the early 17^th-century Jesuit grammar later reprinted as Buelna (1890), notes (p. 44) the existence of nouns ‘caracen de singular’, i.e. ‘lacking singular’, and list several examples familiar from the modern language, e.g. supem ‘el vestido’ [dress(es)], puusim ‘los ojos’ [eye(s)], nacam ‘las orejas’ [ear(s)], as well as a couple that are no longer pluralia tantum, e.g. vatatzim ‘la rana’ [frog(s)], and tzoquim ‘las estrellas’ [star(s)].

The only suggestion of a generalization concerning this grammatical class that we have seen in previous literature involves the frequent application of the plural suffix to borrowed words. Molina et al. (1999: -m), list one sense of the plural suffix -m as a marker of borrowed Spanish terms:

“3. marker of terms introduced by Spanish culture (especially tools; ex. livrom,

‘book/books’; martiom, ‘hammer/hammers’; tisiriam, ‘scissors’)”

In a similar vein, Estrada Fernández and Guerrero (2007: 3) suggest that Spanish bare nouns are generally borrowed with a plural or collective interpretation and hence receive the -m suffix.⁵ It certainly is the case that many Spanish borrowings have become pluralia tantum nouns in Hiaki (8a–g). However, many have not (e.g. 8h–n):

We return to discussion of the PT status of borrowed nouns in section 3.1 below.

We undertook the first systematic review of PT nouns in the language by examining all the nouns in the Molina et al. (1999) dictionary, supplemented with forms and examples from the Estrada Fernández et al. (2004) dictionary. Of the 1421 noun headwords in the Molina et al. dictionary, 316 are given with the plural suffix. In other words, 22% of the headwords in the dictionary seem to be pluralia tantum, in contrast to previous assessments that suggested “there are only a few inherently plural nouns” (Dedrick & Casad 1999: 131). For a complete list, see the DOI linked under “Data Availability”.

While our starting point was identifying these headwords, where possible we also looked at existing example sentences using these terms and at times elicited new ones. Occasionally, we found a difference between dictionary example sentences and elicited speech. For example, both dictionaries only listed the PT form veho’orim ‘lizard(s)’. Estrada Fernandez et al. (2004) give the following example, in which veho’orim is marked for plural but receives a singular translation:

However, one speaker of Arizona Hiaki identified veho’ori ‘lizard’ as an acceptable singular form. We conclude that there can be lexical variation between speakers concerning the PT status of particular nouns.

Given the standard observation that pluralia tantum nouns are often drawn from recognizable semantic categories (see, e.g. Williams 1994: 12), we investigated whether we could make any plausible generalizations about which categories of referents are marked as pluralia tantum in Hiaki. This process yielded immediate and robust results, with almost all Hiaki PT nouns falling into one of several clear semantic categories.⁷ We present the categories we found in (10), in an order suggested by Grimm (2018: 543)’s scalar organization of entity types exhibiting collective/singulative marking in four unrelated languages.

It is important to note that the connection between being a PT noun and denoting in a relevant semantic category is a one-way implication: If a Hiaki noun is a PT noun, it belongs to one of these semantic categories. However, there are nouns that denote in these categories which are not grammatically PT. Consider, e.g. kafe, ‘coffee’ and ohvo, ‘blood’ which denote liquids but are formally singular, or choki tachiria ‘starlight’, which denotes an extensionally cumulative concept, but is formally singular. Such cases play an important role in the featural analysis we propose in section 4.2 below.

2.3 Hiaki PT nouns in a cross-linguistic context

We see in this categorization of Hiaki PT nouns some similarities with PT behavior in other languages. Koptjevskaya-Tamm & Wälchli (2001: 630), for example, note that substances, complex artifacts (i.e., “multipart objects”), diseases, environments, festivities, periods of time, and activities with multiple participants are all common categories for PT nouns in the circum-Baltic languages. And we do see PT examples of each of these in Hiaki. Indeed, with a little introspection and attention to literature discussing English (e.g., Wierzbicka 1988; Williams 1994; Acquaviva 2008), we can identify sets of English PT nouns which instantiate similar categories:⁹

Considering this data in a cross-linguistic context, we see that what stands out about Hiaki overall is not the particular categories of PT nouns, but rather the size of the PT inventory. Other language families that have been reported to have a ‘large number’ of PT are Baltic, Slavic, and Finnic but it is unclear what ‘large’ means; studies (Koptjevskaya-Tamm & Wälchli 2001; Corbett 2019) look at item lists of a few dozen rather than at the whole nominal inventory as represented in the dictionary. The 22% of Hiaki nouns that are PT seems quite noteworthy in this context.

From a cross-linguistic perspective, it is also worth noting that PT nouns are unusual in the Uto-Aztecan language family (p.c., Friends of Uto-Aztecan Conference 2021, p.c. Kenneth Hill), despite being very robust in Hiaki. This is perhaps surprising given that in languages where PT is well developed, it tends to be diachronically stable (Koptjevskaya-Tamm & Wälchli 2001: 632), and is reconstructed to the proto-language. Although we know that Hiaki PT were documented in the early 1700s, and may thus be diachronically stable, it is surprising that there aren’t robust pluralia tantum nominal systems elsewhere in the language family. This aspect of the Hiaki system certainly deserves further investigation within the context of the development of the language family.

3.1 Generality and productivity of PT categories in Hiaki: Borrowed Ns

The key observation that emerges from our data is that the Spanish borrowings which are marked as PT fall into the same semantic categories as native Hiaki PT. All the PT borrowings in Molina et al. (1999) fall into one of the relevant semantic categories. We illustrate with some salient examples below:

Spanish borrowings that do not denote in a PT-type semantic category are typically not borrowed as PT nouns (13), making it unlikely that grammaticization as a PT noun was a nativization strategy for borrowings.¹⁰ This point is underscored by the fact that most of these words have been fully nativized phonologically and morphologically, and some are not salient as borrowed terms to speakers. For example, kava’i ‘horse’ has undergone considerable phonological change from the Spanish original caballo; as have kus ‘cross’ (from Spanish cruz), and kompae (from Spanish compadre).

As noted above, a few borrowings which do clearly denote in one of our PT domains have failed to be grammaticized into a PT category in Hiaki. We list every such case that we have been able to identify with certainty below:

In sum, we have seen that the many borrowed words which have been nativized as pluralia tantum nouns fall into the same categories as native PT in Hiaki. We thus claim that their participation in the PT system is largely (or entirely) driven by their semantic content, rather than by their status as borrowings. We observe that one of the reasons that the use of PT marking with borrowed nouns is particularly salient is because two of the relevant semantic domains are areas that saw a lot of borrowing upon contact with colonizing Spanish culture. This includes names for hand tools, where native Hiaki terms like sena’asom ‘pascola’s disk rattle(s)’ or hi’ikiam ‘needle(s)’ were presumably already PT before contact, and names for articles of clothing and fabrics, like the native Hiaki terms hipetam ‘mat(s)’ or sekawam ‘matachin crown ribbon(s)’.

In the following subsection we look more closely at specific categories of PT that might be surprising from an Indo-European perspective: hand tools, constellations and times.

3.2 Unusual pluralia tantum categories: Hand tools, constellations and times

Hand tools, at first glance, may seem to be a surprising PT category: many are prototypically only used one at a time (e.g. paalam ‘shovel’ and mache’etam ‘machete’). Hand tools are not typical pluralia tantum nouns in Indo-European languages, where most extant studies on PT have focused. The fact that the behavior of these nouns is unexpected from an Indo-European perspective may have even motivated the earlier proposals that PT marking was a borrowing strategy in Hiaki. However, as mentioned above, native Hiaki terms for handheld items are also generally pluralia tantum, even when they are only used in one hand:

Further, the vast majority of Spanish borrowings for hand tools are mapped to the PT category—and as we would expect for this semantic domain, there are many such examples:

Why should this semantic domain in particular show such robust PT categorization? We hypothesize that tools that are held in the hand are categorized as PT because hands are—mamam ‘hand(s)’, *mama ‘hand’. Whether discussing one hand or many, as for all paired body parts in the language, the plural form is always used. The categorization of paired body parts as PT is cross-linguistically unexceptional. What we propose (based on an idea from Jacqueline Guéron, p.c.) is that tools are, conceptually, extensions of the hand. Since hands are canonically PT, so too are tools.¹¹ This would be an interesting domain for typological exploration—as far as we know, most tools that are marked as PT in Indo-European languages have bipartite internal structure (scissors, pliers, binoculars), and it is that internal structure that underlies their membership in the PT category; Hiaki is the first language we have heard of where even non-internally-complex hand tools like shovels are categorized as PT.¹²

Another surprising PT category (from an Indo-European perspective) was the names of constellations, which are PT even when the image and word associated with the constellation is not. For example, one of the names for the constellation referred to in English as ‘The Big Dipper’ is kaarom, a PT word derived from the non-PT Hiaki kaaro meaning ‘car’ or ‘cart’ (itself a Spanish borrowing, from carro ‘cart’) combined with the plural marker -m. The constellation, crucially, is not being described as containing multiple cars or carts; instead, it is the multi-part nature of a constellation—a configuration of multiple stars—that gives it its pluralia tantum status. Similarly, the alternative native Hiaki name for the Big Dipper, choki araum, lit. ‘star plow-pl’, is clearly iconically based on the overall shape of the constellation configuration, rather than on some concept involving multiple plows.

Finally, it is worth mentioning that Hiaki names for specific times in a 12-hour clock are PT as well, and exhibit a few interesting idiosyncracies. Most such time names use native Hiaki numbers, e.g. woi-m-po two-pl-at ‘at two o’clock’, and are unsurprisingly marked plural. However, ‘one o’clock’ must use the Spanish word for one, (uno) as the base, forbidding the use of either of the native Hiaki words for ‘one’, senu or wepulai, and furthermore uno- must be marked for plural: (h)uno-m-po ‘at one o’clock’. It’s on the basis of this last example that we assume time number names represent a type of abstract PT noun.

4.1 Number marking in Distributed Morphology

For concreteness, we adopt the Distributed Morphology framework for modeling the morphosyntactic interface. In Distributed Morphology, morphosyntactic features occupy the terminal nodes of a (Minimalist) syntactic tree structure (Halle & Marantz 1993; Siddiqi 2010, a.o.). These features receive phonetic expression via competition by different Vocabulary Items (‘morphs’) when the derivation reaches PF (‘Phonological Form’), and they receive semantic interpretation when the derivation reaches LF (‘Logical Form’). The framework is designed to support full and explicit mappings of morphological representations to semantic representations via the syntax. In the usual case, these representations will stand in a regular one-to-one relationship. For example, in the case of a regular Hiaki count noun, which occupies the bottommost element of a syntactic nominal functional projection containing at least NP and Num,¹⁴ the featural content of the Num head determines both whether that count noun is interpreted as singular (atomic) or plural (non-atomic) on the LF side, and whether that count noun is marked with -∅ or -(i)m on the PF side.

Challenges arise, however, when we consider pluralia tantum count nouns, which we have seen are always marked plural even though they can denote in both the atomic and non-atomic domains. We repeat the data from (7) above in (17). Recall that supem ‘dress’ (in 17a) and tahkaim ‘tortilla’ (in 17c) always bear plural inflection, although they can refer to singular or plural dresses/tortillas without issue:

We could model this by hypothesizing that a PT noun like tahkaim or supem bears an uninterpretable [–singular] feature, which is always realized as -(i)m and is accordingly active in the syntax in concord operations, but we then would need to explain why this feature doesn’t receive a [–singular] interpretation at LF, the way a regular [–singular] Num head must. In short, why do PT nouns allow this mismatch, where their syntactic number marking and semantic number interpretation are contradictory?

This challenge is compounded by the fact illustrated in (3) above and repeated below as (18) that although syntactically controlled number marking always shows plural agreement with a PT noun regardless of semantic number, number-sensitive suppletive verbs reflect the semantic/interpreted number of their Theme argument,¹⁵ and do not reflect its morphological plural number marking.

That is, interpreted number and actual syntactic number marking must be modeled independently of each other. We turn to our proposal for how to do this in DM in the next section.

4.2 Options for modeling PT number independently of semantic number in DM

To model this chimeric system, we will adopt and adapt an insight from Weschsler and Zlatić (2000)’s treatment of Bosnian/Serbian/Montenegrin/Croatian agreement in Head-driven Phrase Structure Grammar, according to which purely morphological phi-features, called ‘Concord’ features, are distinct from semantically interpreted phi-features, called ‘Index’ features. Nominals bear both Index and Concord features, and different classes of agreement phenomena are sensitive to the distinct types of features. This allows an account of ‘mixed’ agreement patterns within the extended nominal projection of Bosnian/Serbian/Montnegrin/Croatian. (See also Landau 2016 for a Distributed Morphology analysis of a Hebrew PT nominal exploiting Wechsler and Zlatić’s feature types).

Plural marking on pluralia tantum nouns, adjectives and determiners in Hiaki is clearly a case of uninterpretable Concord features controlling number agreement throughout the nominal domain. In contrast, we claim that the number-sensitive suppletive verbs are conditioned by the interpretable Index features of their Theme argument. The verbs pay no attention to the Concord features. Unlike the proposal in e.g., Wechsler and Hahm (2011: 259–260) for treating interpretable number marking on the target predicate as semantically active just in case number on the NP [=DP] is underspecified, we locate all semantically interpretable number in the DP itself. In our analysis, predicate number marking is never semantically active, but is conditioned by the syntactic Num features which are semantically active, namely Index features. We explicate this distinction between the approaches further below.

We suppose that in the unmarked case, on a regular non-PT count noun, the Concord feature copies the value of the Index feature and thus exhibits uniform behavior across both nominal and verbal number markers—plural-marked nouns co-occur with plural suppletive verbs and vice versa. In the case of countable PT nouns, however, the values for Concord and Index number can come apart. We analyze each of the four logically possible combinations from Table 2 in turn below, which we summarize again to remind the reader of the categories we identified:

To begin, we consider the case of a ‘normal’ count noun, where morphological number marking in the DP perfectly reflects number interpretation throughout the clause, as in the case of human-denoting nouns like bwikreo ‘singer’.

Here, the values of the Concord and Index features in Num line up perfectly.¹⁶ We hypothesize that for such ‘normal’ count nouns, the Concord features of Num are underspecified and their values are filled in by a feature-copying operation which copies the interpretable Index feature value in Num. This ensures that in the ‘normal’ case the exponent of Num at PF and the interpretation of Num at LF will match.

The Concord feature in Num then determines the number marking on all elements in the extended nominal projection which bear Concord features. In DM, this is cashed out by a value-copying operation. The Concord feature on e.g. D enters into an Agree relationship with the Concord feature on Num (via Agree Closest, as in e.g. Bošković 2009) and copy its value. (We indicate pre-specified Concord (C:) features in bold font, and copied Concord features in greyscale font. Index features are given simply as ‘I:[±sg]’).

We turn now to the mechanics that determine the insertion of the correct verb form. We implement suppletive verb agreement in DM via vocabulary insertion rules of the form in (21) below, as argued in Harley (2014). (See Harley et al. (2016) and Bobaljik & Harley (2017) for detailed discussion and motivation for using the plural form as ‘Elsewhere’, rather than the singular.) In the case of a ‘normal’ count noun, of course, we cannot be sure which of the features—Index or Concord—is conditioning the verbal form, since they have identical values, but to anticipate, we include the “I:[+sg]” notation in the conditioning context in the rule, since it will become crucial momentarily:

To repeat: when a normal count noun is the Theme argument of a suppletive number-sensitive verb, as in (21), the form that realizes the root matches both sets of features on the noun, so we can’t disentangle the question of whether the verb is sensitive to Index or Concord features with this class of nouns.

Let us now turn to the case of countable pluralia tantum nouns like supem ‘skirt(s)’ or paalam ‘shovel(s)’, whose number marking within the DP is entirely and always plural, but for which the verb changes form depending on whether singular reference or plural reference is intended, and where combination with the singular numeral senu/wepulai ‘one’ is grammatical despite the morphological plural marking (as in ex. (7a/17a)).

With such PT nouns, the nominal form requires a plural Concord feature in its Num head. We can implement this within DM with a spell-out rule for the nominal that allows insertion only into structures containing a Num head with value C: [–sg], as illustrated in (22a) below. The Index feature of the Num head, however, is not restricted at all, and can vary between singular and plural according to the speaker’s intention, i.e., it is compatible with both I:[+sg] and I:[–sg]. The resulting structure is illustrated in (22b):

(Note that this morphosyntactic structure has the desired effect of representing the -m suffix as a normal exponent of a [–sg] Num head, rather than as a substring of the noun itself. This is necessary to account for the absence of the plural suffix on PT nouns in compounds, as in (6b, d) above.)

To reiterate, while the Index feature on a PT noun varies according to the intention of the speaker, the Concord feature is restricted to only [–sg], and that value is copied to all other locations that probe for Concord number features.

Now we see why it was crucial to include the I:[+sg] specification in the conditioning context for suppletive verb insertion in the rule in (21) above: The verb form of a suppletive verb with a PT Theme noun is conditioned by the value of the Theme’s Index feature, rather than by its Concord feature, since the verb form varies according to the intended semantic number.

Now let us turn to uncountable nouns, i.e. mass nouns. We consider first the cases like kafe ‘coffee’ or ohvo ‘blood’ which are marked singular in the nominal domain: uu kafe ‘the.sg coffee’, not *ume kafem. As expected, such mass nouns condition singular verb forms for suppletive number-marking verbs (like English mass nouns):

Since such nouns are ungrammatical with plural nominal morphology and take singular verb forms, we can conclude that both their Concord and Index features are marked [+sg]. Indeed, these forms are crucial to our choice to use the Farkas and de Swart (2010) [±sg] feature notation in our analysis of Hiaki number, rather than the [±atomic] number feature often proposed elsewhere in the semantic literature on number (see, e.g. Harbour 2014). These nouns pattern with singulars but cannot be characterized as having a [+atomic] feature, since they denote in the non-atomic domain.

The [+sg] Concord feature on singular mass nouns might arise from the copying of a [+sg] Index feature into an underspecified C: feature slot (as it does for regular count nouns), or it might be required by a nominal vocabulary item rule (as for the plural concord exhibited by countable PT nouns). We will adopt the former idea that it is specified by the copying rule, since the (conceptual) non-pluralizability of mass nouns seems adequate to prevent them from accidentally appearing in the plural,¹⁷ but nothing hinges on this choice.

Since the Index feature conditions verb insertion, the model predicts that the verb form of number-conditioned suppletive verbs with these mass nouns as their Theme will necessarily be singular, as indeed it is.

Finally, we turn to the (large) class of noncountable pluralia tantum mass nouns like vaa’am, ‘water’. Interestingly, like kafe, these condition singular verb forms:

This behavior entails that they have a [+sg] Index feature in our model. They also must morphosyntactically require a [–sg] Concord feature, as for countable PT nouns, which we implement with the same kind of vocabulary-item spell-out rule, given in (26a). The resulting structure is illustrated in the tree in (26b) below:

The observation that the semantically driven agreement indicated by the verb stem is singular even with PT mass nouns confirms our conclusion above: the mass/count distinction in the Hiaki number system cannot be captured using a [±atomic] feature interpreted as such, because the denotation of such non-countable nouns is clearly not [+atomic]. The Hiaki facts require us to create a model where mass nouns and singular count nouns pattern together at the Index feature level, but not at the Concord feature level.

With this class of mass nouns, then, we see a surprising result. Their Concord features pattern with those of plurals, which is consistent with their position at the bottom of the Scale of Individuation, but their Index features, as revealed by suppletive verbal number agreement, are singular. That is, in the features that one would think ought to pattern most closely with their conceptual ‘non-individuated’ status, they reveal themselves to be singular. It is only in the Concord morphology, which does not bear on their actual interpretive individuation, that we see the patterning suggested by the Scale of Individuation.

It is worth remarking that this behavior is significantly different from that of English, which uses plural marking to indicate number-neutral denotations for count nouns (Did you buy boxes/*box?), but in which mass nouns, morphosyntactically similar to plurals in being able to occur without an overt determiner (I bought a box/boxes/wine), in fact control singular agreement with demonstratives and the finite verb (Wine is/*are delicious; That/*those wine is delicious). In applying our model to English, then, mass nouns would seem to be represented with [–singular] Index features but [+singular] Concord features, precisely the opposite of the Hiaki situation.

5.1 Hiaki pluralia tantum, verbal suppletion & Corbett’s Agreement Hierarchy

In Table 1 above, we illustrated examples from the extreme ends of Corbett’s typological survey of agreement patterns, which led him to propose the Agreement Hierarchy. According to the AH, if a nominal controls agreement on a target higher in the hierarchy (farther to the right of Table 1), it will invariably control agreement on targets lower in the hierarchy as well (farther to the left on Table 1). So, e.g., plural anaphoric agreement with a given nominal predicts that said nominal will control plural verb agreement and plural nominal concord, since verb agreement and nominal concord are lower on the hierarchy than anaphoric agreement. In contrast, mandatorily plural nominal concord could in principle co-occur with singular verb agreement and singular anaphoric agreement, since they are higher on the hierarchy. Importantly, according to the AH, mandatory verb agreement predicts matching concord agreement, but says nothing about the nature of anaphoric agreement – typologically speaking, nouns exist that look plural, and trigger plural verb agreement, but which allow for singular anaphoric reference when semantically appropriate. Anaphoric agreement is the highest agreement type in the hierarchy.

As noted above, the Hiaki facts pose a challenge to Corbett’s Agreement Hierarchy at first glance. Hiaki PT nouns trigger concord on determiners and adjectives (27a), as we have seen, and they are referred to with plural anaphors in subsequent discourse (27b):

The Agreement Hierarchy, which places verbal agreement below anaphoric agreement, thus predicts that in Hiaki, verbal agreement should reflect the formally plural status of pluralia tantum nouns.

However, as we have seen, verbs in fact agree according to the actual semantic number of the referent—the value of the Index feature—not according to the formal Concord plural feature (28):

This looks prima facie like a problem for the Agreement Hierarchy as stated—we see PT concord agreement and PT anaphoric agreement, but not PT verbal agreement.

We could conclude from this that the hierarchy is different for Hiaki—maybe verbal agreement is exceptionally higher than anaphoric agreement for this language—but in fact, we propose to take a different tack, and argue (with Harley et al. 2016 and Bobaljik & Harley 2017) that Hiaki verbal ‘agreement’ is not normal agreement, i.e., it is not syntactically mediated via an Agree relation. Only syntactically mediated agreement is subject to the generalizations of the Agreement Hierarchy.¹⁸

5.2 Hiaki pluralia tantum and Grimm’s Scale of Individuation

We have seen above that the particular nouns which exhibit PT marking fall into recognizable semantic clusters, corresponding to Grimm (2018: 543)’s scale of ‘entity types’. As with PT nouns in many languages, it seems intuitively correct to assert that these categories generally have a kind of notional ‘natural’ plurality to them: PT body parts occur in pairs, constellations are made up of many stars, aggregate substances like rice or beans are made up of many individual grains, etc. These categories, interestingly, correlate with the semantic categories of nouns that exhibit marked singulars—singulatives—in languages that have them, as demonstrated by Grimm (2018).

On the basis of the behavior of marked singular nominals cross-linguistically, Grimm (2018) proposed that nominal referents can be categorized on a ‘Scale of Individuation’, according the “propensity for the entity described by the noun to occur as an individual”, which he connects to the ‘accessibility’ of the unit interpretation for any given lexical nominal. Marked singulars are nominals that show overt number marking only when they are used with singular reference, rather than showing overt marking when they are used with plural reference, as is the case in languages like, e.g., English. In a typological investigation including Welsh, Maltese, Breton and Dagaare, he showed that marked singulars tend to occur in semantic categories that exhibit conceptual plurality.

Considering the particular semantic content of the nouns which exhibit marked singulars in all these languages, he proposed a conceptual Scale of Individuation, as follows (Grimm 2018: 549, based on a notion in Comrie 1989: 199):

As usual with such hierarchies, this one is intended to describe the existence of an implicational relationship: Grimm describes this as follows:

“If two individuation types, say granular aggregates and collective aggregates, stand in a less-than-or-equal-to relationship on the scale of individuation, they will map to categories in the language’s grammatical-class inventory that preserve the function.”

(Grimm 2018: 550)

This leads us to ask whether the Hiaki PT categories can be insightfully modelled using the Scale of Individuation. Although categories of Hiaki referents that are encoded with PT nominals correspond almost exactly to Grimm’s ‘entity types’ scale, they do not form to a coherent proper subset of individuation types as described by Grimm (2018: 550). The PT nominals of Hiaki span the whole Scale of Individuation, as do the singular nominals. In Hiaki, there exist singular mass nouns, where no unit reference is possible, which are nonetheless morphosyntactically singular, as in ohvo ‘blood’ and kafe ‘coffee’. In contrast, we also see countable PT nouns whose referents have a high degree of perceptual individuation, and for which unit reference is available and even likely (see section 6.3 below), like mache’etam ‘machete’ or hi’ikiam ‘needle’. Indeed, the very countability of PTs—the availability of unit reference as exemplified by quantification by ‘one’—means that (lack of) individuation is precisely not criterial for membership in this ‘plural’ lexical morphological category.

Overall, we conclude that Hiaki pluralia tantum nouns are more usefully characterized by Grimm’s set of entity types than by the Scale of Individuation. Nonetheless, as mentioned above, there is a clear sense in which most of the entity types whose corresponding nominals are in the countable pluralia tantum category do exhibit some kind of ‘notional plurality’ – a group of stars, pairs of body parts, etc. This leads us to ask whether another line of thinking about morphological markedness from the literature could be useful in accounting for the distribution of pluralia tantum nominals in Hiaki. Can we appeal to a frequentist notion of ‘plural reference dominance’ to account for which nouns acquire ‘always plural’ status, morphologically speaking? The Hiaki case presents several features which bear on the debates about predictability and morphological marking in potentially interesting ways.

5.3 ‘Frequentist’ view of morphological markedness in Hiaki PT nouns

The concept of ‘plural reference dominance’ for marked singular categories is construed in terms of predictability of form. On one formulation of this idea, predictability is straightforwardly determined by the frequency of occurrence of a particular form-meaning correspondence in the input (Haspelmath & Karjus 2017). This builds on an interpretation of morphological markedness that is itself dependent on the usual interpretation of Zipf’s Law of Abbreviation (Zipf 1949), encoding the observation that word frequency is inversely correlated with word length: the less frequently a word is used, the ‘bigger’ it is, morphophonologically speaking. Intuitively, if the more frequent form is shorter than the less frequent form, the system minimizes speaker effort while maximizing hearer interpretability: A hearer is much more likely to be able to correctly identify a highly frequent word from a smaller phonological signal than they are to be able to identify a less frequent word from a smaller phonological signal. Therefore, less frequent words should be represented with more phonological material to allow a hearer to unambiguously identify them; conversely, highly frequent words should be represented with less phonological material to minimize speaker effort where it is not needed. Haspelmath and Karjus (2017) argue that ‘uniplex-prominent’ nominals—nouns which are more frequently used in the singular—will therefore show overt coding for plurality, which they call ‘plurative’, while ‘multiplex-prominent’ nominals, which are more frequently used in plural forms, may evolve to show overt coding for singulative.²¹

This general idea about how morphophonological markedness correlates inversely with frequency has recurred in multiple literatures in various guises. Horn (1984), concerned with scalar implicature, dubbed this tendency the ‘division of pragmatic labor’: The morphologically marked member of a pair of inflected forms should correspond to the less frequent meaning. Haspelmath et al. (2014) and Haspelmath (2021) dub this the ‘form-frequency correspondence hypothesis’.

Mattausch (2007) computationally implements a stochastic OT model which shows that ‘bidirectional optimization’—optimization for both hearer and speaker over multiple iterations in a communicative computational model—produces exactly this effect: the morphologically more marked member of a given lexical item with a two-way inflectional contrast is mapped to the less frequent meaning, and the morphologically unmarked member of the pair is mapped to the more frequent meaning.

Haspemath & Karjus (2017) point to the textual frequency of plural forms of translation-equivalent nouns in large corpora in English, Estonian, Latvian, Norwegian and Russian to make the case that singulative forms (in languages that have them) appear with lexical items which tend to be used in the plural more often than in the singular. They find that 18 ‘singulative-prominent’ nominal meanings identified through typological examination of languages with singulative marking are statistically more likely to be used in the plural than in the singular in these large corpora, in contrast with 18 randomly-selected nominals which turn out to be statistically more likely to be used in the singular. Haspelmath & Karjus (2017: 1227) propose a diachronic explanation for the development of singulative forms that is precisely in line with Mattausch’s (2007)’s demonstration of a frequency-based evolution of Horn’s (1984) ‘division of pragmatic labor’. They write “…the correspondence between form and frequency is implemented by diachronic mechanisms which tend to make frequent forms short, because frequent forms are more predictable than rare forms. Ultimately, it is thus predictability that lies at the root of the length difference and the coding asymmetry.”

With this background, we can ask whether this line of explanation can offer insight into the development of the coding of particular Hiaki count nouns as pluralia tantum. The pluralia tantum nouns we have documented above tend to belong to semantic categories that are less individuated, yet are morphologically overtly marked as ‘plural’. This means that the usual Zipfian idea, where less-marked forms should denote plural for nouns with more frequent plural interpretations, does not apply in this case. However, we could imagine a variation on the frequentist idea according to which the grammaticization of plural marking on pluralia tantum nouns could be driven by highly frequent plural uses for these nouns. That is, perhaps the grammatical requirement to be always plural developed out of being used more frequently in the plural than in the singular.

This line of thinking as a possible explanation for why the Hiaki pluralia tantum categories are low on the Scale of Individuation runs into trouble, however, in several domains. We cannot say, for example, that the borrowed Spanish pluralia tantum nouns like paalam ‘shovel’ went through several generations of plural dominant usage and thus finished up as morphologically PT, since they do not have a long history in the language. Or consider lakim ‘lock’, which was likely borrowed directly from English; extensive contact between Hiaki and English did not occur prior to the 1880s, to our knowledge. Another challenge arises in that although mass nouns are extremely low on the scale of individuation, they are not relevantly ‘plural’, as we saw above from the fact that they condition singular forms of suppletive plural-agreement verbs (section 5.2 above); if anything, we would expect that a frequentist motivation for PT coding would result in singular forms for mass nouns, rather than plural forms.

We can also test the ‘plural-reference dominance’ approach by adopting Haspelmath & Karjus (2017)’s methodology of quantifying plural-reference dominance for translation equivalents of individual words in a large corpus of another language, making the assumption that these nouns are used with plural reference in that language at reasonably analogous rates in Hiaki.

To arrive at an initial picture of the plausibility of a frequentist view of PT marking on a few individual nouns in Hiaki, we used COCA (the Corpus of Contemporary American English, Davies 2008) to quantify singular and plural frequences for the nouns wrist, ankle, crowbar and machete, using a proportional sample method.²² Given a string such as CROWBAR,²³ we would like to know what proportion of the 758 occurrences of that string in COCA are true singular nouns (i.e. eliminating all instances of CROWBAR that are the left-hand member of compounds, present-tense verbs, etc.). We arbitrarily selected 100 matches for CROWBAR in COCA and hand-coded each according to whether they were true singulars or not. In our randomly-selected set²⁴ of CROWBAR tokens, 92/100 instances were true singular nouns. Assuming that our sample is big enough to be representative of the distribution of singular nominal uses of CROWBAR throughout the corpus, we extrapolate that percentage and arrive at a count of 0.92*758 = ~697 tokens of CROWBAR as a singular noun in COCA. Now, to compare that number to the number of true plural tokens of CROWBARS we repeat the process, examining 100 randomly-selected instances of CROWBARS to quantify the percentage of tokens in that sample that are true plural nominals. For the string CROWBARS, there were only 86 tokens total, and 86/86 of these tokens were actual plural nominal uses, i.e. 100%. We thus have a singular:plural ratio for the lemma crowbar of 697:86—heavily biased in favor of singular uses.

Repeating this procedure for MACHETE yielded 91/100 singular nouns in our randomly chosen sample of 100 tokens of uninflected MACHETE.²⁵ Thus, multiplying 0.91 times the 1171 total occurences yields ~1066 tokens of MACHETE as a singular count noun. We then searched for MACHETES and examined a randomly chosen sample of 100 hits²⁶ for that string; 100/100 of these hits were plural nouns, so we estimate that all 498 hits for MACHETES in COCA are true plural uses. Thus, we end up with a sg:pl ratio of 1066:498 for the machete lemma, also heavily biased in favor of the singular form.

Taking two other sample words from a different semantic subgroup of our Hiaki pluralia tantum nouns, we compared sg:pl ratios for WRIST and ANKLE using the same methodology. First, for uninflected WRIST, 66/100 tokens in our sample²⁷ were singular nouns; thus we conclude that approximately ~7731 of the 11,714 tokens of WRIST in COCA are likely to be true singulars. Turning to WRISTS, 100/100 tokens in our sample of WRISTS²⁸ were true plurals. There were 4585 total occurrences of WRISTS, so we can compute a sg:pl ratio of 7731:4585 for the wrist lemma, again showing a notable bias towards the singular use.

Finally, for singular ANKLE, 63/100 instances in our set of 100 tokens were true singulars.²⁹ Multiplying this percentage by the total number of occurrences of ANKLE yields 0.63*10,394 = ~ 6548 tokens of true singular ANKLE in the corpus. For ANKLES, 100/100 examples³⁰ were true plural nominals, and there were 4799 tokens total; we thus obtain a sg:pl ratio of 6548:4799, again, showing a singular bias. We summarize these results in the bar chart in Figure 1.

Figure 1: The ratio of approximate singular tokens to plural tokens for four selected English words in COCA whose Hiaki translations are pluralia tantum nouns.

For at least these four words, then, it seems quite unlikely that frequency-based plural-reference-dominance could have driven grammaticalization into the PT category.

Instead of a frequency-driven process yielding pluralia tantum behavior for individual nominals, then, we conclude that Hiaki PT behavior is better thought of as reflecting membership in a particular semantic class, supporting Kurumada & Grimm 2019’s results showing that grammatical coding is more likely to be causally driven by conceptual categorization, rather than by simple token frequency. The particular categories relevant for PT coding in Hiaki, we assume, were already in place as a feature of Hiaki grammar by the time large-scale borrowing of Spanish began, since as we noted above they were documented in the early 17^th century Jesuit grammar (Buelna 1890: 44). As Spanish nouns were borrowed, they must then have been slotted into PT or non-PT categories and ‘nativized’ according to their denotata’s membership in the relevant semantic class.

This notion of grammaticization based on semantic category fits with the overall insight of the Concord/Index system for number features proposed by Wechsler & Zlatić (2000; 2003)—the Concord features are ‘grammatical’ properties, whereas the Index features reflect actual numerosity. As we have seen in the case of pluralia tantum mass nouns, it’s the grammatical Concord features, not the semantic Index features, that are involved in marking a mass noun as a plurale tantum noun. The idea that pluralia tantum status is like a noun class feature within Hiaki connected to membership in some language-specific semantic category is consistent with Corbett’s notion of noun class features in general being grounded in a nonlinguistic semantic category (gender, shape), which is then extended to large groups of nouns when grammaticized (Corbett 2013).

Finally, when trying to reverse-engineer hypotheses about the sources of PT coding in Hiaki, it is worth reemphasizing that PT behavior is not a common property in related Uto-Aztecan languages (p.c. Ken Hill, FOUA 2021 attendees). We assume that this indicates that it is an innovation in Hiaki (and closely-related Mayo), but the path of historical development of this robust pattern seems even more mysterious given that it is unusual for the language family.³¹ We hope that future work can shed more light on the diachronic source of this categorial behavior.