1 Introduction

Since Abney (1987), and especially in the 21^st century, the vast majority of work in mainstream generative grammar has adopted the “DP analysis,” the idea that the syntactic head of the nominal domain is D (not N, as traditionally assumed). However, recent years have seen a resurgence in interest in the traditional NP analysis (Bruening et al. 2018; Bruening 2020; Blümel 2024; also Chomsky 2019: 51; and contributions to Blümel & Holler 2022). Of course, there has also always been resistance to the DP hypothesis from outside mainstream generative grammar.

There have been a number of responses to this development, especially to Bruening et al. (2018), Bruening (2020). Preminger (2020) reiterates an argument for NumP from modern Hebrew originally from Ritter (1991), and Barrie et al. (2021) give numerous arguments for DP from over a dozen typologically diverse languages. Salzmann (2022) gives a somewhat more cautious defense.

Here, I want to mobilize a fairly simple slice of a complex topic in English syntax (discussed extensively in DP-analytic terms by Heycock & Zamparelli 2005) in defense of functional structure in the nominal domain. The argument specifically centers on coordination of bare count nouns, particularly the behavior of these constructions with different kinds of determiners. In short, noun-noun coordinate structures seem to be singular when they occur with singular universal quantifiers each and every, as in (1b), but they seem to be plural when they occur with other determiners, like our in (1a).

My core claim is that, in an adequate account of these facts, the noun-noun coordination itself cannot be the controller of morphosyntactic number for the full nominal projection. In section 2, I show that noun-noun coordinations cannot be singular. However, in section 3, I show that they are not plural, either. Section 4 argues that these facts force an analysis where the number feature of the full nominal projection is external to the noun-noun coordination. Section 5 sketches such an analysis, with much in common with Heycock & Zamparelli’s (2005) account of many of the same facts.¹ Section 6 concludes.

2 Noun-Noun Coordination

In English, when two singular count nouns are coordinated, the resultant Noun-Noun Coordination (NNC) typically behaves as a plural noun. That NNCs are morphosyntactically plural can be seen conclusively in plural agreement, as in (2)/(3). The NNCs pattern with plural nouns in licensing plural agreement on the verb, and in many cases not licensing singular.

The ‘#’ diacritic in (2b)/(3b) explicitly recognizes that these forms are grammatical but infelicitous—in many instances, analogous constructions are fine (e.g. “My hope and prayer is…”). I further discuss such cases (briefly) in section 5. What’s crucial, though, is that the full nominal expressions can be plural in cases such as (2b)/(3b); these particular examples, where singular is infelicitous, illustrate this especially well.

Diagnostics for semantic/notional plurality also identify NNCs as plural. For example, collective predicates cannot have singular count nouns as their subjects (collective nouns like team or herd are of course exempt from this generalization). Any of the individual subject nouns in (2)/(3) would fail to be the subject of a collective predicate, but there is no problem with the NNCs in (4)/(5). This suggests that the NNC subjects can be semantically plural.

Anaphoric reference also suggests that NNCs are semantically plural. Plural pronouns, for example, cannot take singular NPs as their antecedents (again, the generalization doesn’t apply to collective nouns), but it is fine for them to take NNCs as their antecedents.²

Thus, the distribution of NNCs resembles that of plural nouns.

3 NNCs with singular universal quantifiers

However, there are also ways in which NNCs cannot be plural. Specifically, NNCs formed from singular count nouns can be used with singular universal quantifiers (SUQs) each and every.³ SUQs are not compatible with morphosyntactically plural nouns, yet they work fine with NNCs, and license singular agreement with them.

It is important to clarify that, although (8b)/(9b) show singular agreement, the agreement facts alone are not what I am highlighting as non-plural behavior of NNCs. Instead, the key property is compatibility with SUQs at all. Coordinated nominals and ordinary plurals form a semantic natural class: both denote mereological sums (Link 1983; Hoeksema 1988; among much other work).⁴ That being the case, the difference between morphosyntactic plurals and NNCs in (8)/(9) is a matter of syntax, not a semantic distinction. It is in this respect that NNCs seem not to be morphosyntactically plural.

That said, subject-verb agreement is relevant in ruling out an analysis of (8b)/(9b) in terms of ellipsis. Specifically, one might seek to explain the compatibility of NNCs with SUQs via an ellipsis analysis, while maintaining that NNCs are fundamentally plural, either with conjunction reduction (10) or with some kind of quantifier deletion (11).

We can reject these elliptical analyses, however, because plural agreement is ungrammatical with these constructions, as (12) shows.

The unavailability of plural agreement in (12) confirms that the NNCs we see in (8b)/(9b) really do involve bona fide coordination of nouns, not of larger categories as in (10)/(11). Specifically, (12) rules out quantifier deletion along the lines of (11). While the hypothetical underlying forms in (11) do indeed allow singular agreement, they also allow plural. In fact, plural is required with these forms as subjects of reciprocal predicates (Winter 1996: 27; Krifka & Modarresi 2024: 659).

If elliptical structures like those in (11) really gave rise to the apparent NNCs in (8b)/(9b), then we would expect to see plural agreement in sentences like (14). In fact, neither singular nor plural is available in such examples.⁵ Apparently, this is because the reciprocal predicates require a plural subject, but the actual subjects here are unambiguously morphosyntactically singular (i.e. not underlain by the elliptical structures shown).

See Heycock & Zamparelli 2005: 254–255 for additional arguments against ellipsis as in (11).

Conjunction reduction along the lines of (10) can also be ruled out, unless there is some way to block the same form of conjunction-reduction applying in cases like (2)–(5) with ordinary determiners. If the same kind of ellipsis were possible with ordinary determiners, we would expect forms like (15) to be grammatical. But they aren’t, a well-known problem for ellipsis analyses of this sort.

Again, I stress that, though the unavailability of plural agreement in (12) is relevant in that it militates against an ellipsis analysis, subject-verb agreement is not the property with respect to which I am pointing out that NNCs seem to be non-plural. Instead it is the fact that morphosyntactic plurals are not compatible with SUQs, while NNCs are (in spite of their semantic similarities), that is crucial.

4 Problems for an NP analysis

There are many questions one could ask about these facts. For example, how should we analyze the compositional semantics of SUQs in light of their behavior with NNCs? Why are both singular and plural agreement allowed with coordination of full QPs headed by SUQs, while only singular is typically allowed when an SUQ occurs with an NNC? These (and many other) questions may be interesting and important (see Heycock & Zamparelli 2005 for relevant discussion), but they are not essential to the narrow point I am making here. Instead, the following formal generalizations are specifically relevant to present concerns (i.e. about the presence/absence of functional structure above N⁰ in the nominal domain).

What I would like to argue now is that it follows that NNCs do not inherently project a morphosyntactic number feature.

To make that point, we need to establish some background assumptions about the mechanics of subject-verb agreement.⁶ First, subject-verb agreement is a purely formal phenomenon, determined by a syntactic dependency between a valuer and a valuee. I say that it is a formal phenomenon, as opposed to a semantic phenomenon, to reflect the assumption that agreement marking on the valuee has no semantic import.⁷ Specifically, I assume that syntactic constituents bear morphosyntactic features, which can be either valued (φ) or unvalued (uφ). The agreement dependency is established via an Agree operation (Deal 2025); the valuee probes its c-command domain and copies the features of the closest valuer, with “closest” understood in terms of Minimal Search (Ke 2019; 2022).

In the case at hand (subject-verb agreement in English), the valuee is T⁰, and the valuer is an element X⁰ in the subject nominal which bears relevant valued φ. The diagram in (17) illustrates schematically (I assume a standard VoiceP analysis following Kratzer 1996).

Thus, the featural properties of the subject (for purposes of agreement) are determined by the featural properties of X⁰.

With this background, the relevant question concerns the nature of X⁰, especially in NNC constructions. An NP analysis would ordinarily predict X⁰ to be N⁰. But of course X⁰ can’t simply be N⁰ in the case of NNCs, because there is not one unique N⁰ involved; an NNC involves two or more N⁰s. Instead, an NP analysis can make the NNC the featural locus in one of two different ways. One option is to say that coordinate structures are inherently plural. For example, perhaps the conjunction and bears an intrinsic plural feature (see Murphy & Puškar 2018: 13; and discussion in Zhang 2009: 53), or maybe coordinate structures are adjuncts to a null pronoun whose features project to the NNC (along the lines of Progovac 1997). Another option is to rely on feature resolution mechanisms of some kind. Such mechanisms are discussed throughout the literature on coordination and agreement (see e.g. Citko 2018: 8–10); they might involve agreement of conjuncts with and, or they might involve feature percolation in some form.

Whichever of these routes the NP analysis takes, we will have a situation where the featural properties of the full nominal projection are determined by those of the NNC, and where those of the NNC are determined by its parts.

And therein lies the rub. In order to capture property (16b), the parts of an NNC must conspire to make the NNC non-plural. SUQs are incompatible with plurals, so NNCs must be non-plural there. But the very same NNCs, with what appears to be the same internal structure, trigger plural subject-verb agreement (16a), suggesting their internal structure makes them plural after all. An NP analysis of these facts therefore cannot be maintained under the assumptions given above.

5 A NumP-based alternative

We seem led to conclude that NNCs do not have a number feature at all. But if that’s the case, then how do we explain their behavior in (2)–(7), where they seem to behave as plurals? There is a natural solution to this apparent problem in positing a functional NumP projection in the nominal domain, such that Num⁰ composes with (bare) NP and D⁰ composes with NumP.

This structure is common in the literature, often with additional functional projections. For example, Heycock & Zamparelli (2005) assume a PlP projection between NumP and NP; for them Pl⁰ handles the semantic singular/plural distinction, while Num⁰ hosts numerals. Here I keep to the simpler (18), as this is all that is needed to make the point at hand.

With this setup, we can assume that singular vs. plural flavors of Num⁰ determine the number features of the full DP. I will further assume that the different versions of Num⁰ have different semantic effects. For concreteness, we can adopt notions of singular and plural from Krifka & Modarresi (2024: 652–653). Their definitions are based around the concept of mereological atoms (Link 1983; Champollion & Krifka 2016): semantic plurals consist of two or more atoms, while semantic singulars do not consist of two or more atoms.⁸ These interpretive differences will be relevant to restricting the compatibility of Num_SG vs. Num_PL with NNCs. (Alternatively we could posit a PlP projection as in Heycock & Zamparelli 2005).

Finally, we assume that SUQs bundle D and Num: they select a bare NP (not pre-qualified by Num) and project a DP with singular features. The intended structure is illustrated in (19a). It could just as well be represented in a Jackendovian way (in the style of Jackendoff 2002) as in (19b).

These fairly elementary analytical assumptions resolve the apparent problem posed by (16). NNCs can be plural with ordinary determiners (16a), because in this case they occur with a plural number projection Num_PL. On the other hand, if they were to occur with Num_SG, it would usually cause a semantic clash. In (20b) for example, there is no natural way for the expression the square and circle to denote a semantically singular entity. Hence NNCs cannot typically be singular (16b).

As a caveat to this, if an NNC could be naturally interpreted as a semantic singular, then nothing stops it from occurring with Num_SG under the right interpretation. This is desirable, given that there are many cases like (21) where exactly such an interpretation is possible, with the expected morphosyntactically singular behavior. See Heycock & Zamparelli 2005 for detailed analysis of this construction.⁹

As for the behavior of SUQs, in fact, on this analysis, SUQs are compatible with neither plurals nor singulars. Instead, they accept only bare NPs (which are innumerate). The reason SUQs appear to be compatible with singulars is that the singular form is syncretic with the bare form. Plurals thus cannot occur with SUQs, assuming the plural -s is a reflex of Num_PL.¹⁰ NNCs on the other hand are perfectly acceptable, assuming coordination of bare NPs yields a bare NP. The diagrams in (22) illustrate the proposed analysis of SUQs. (22a) and (22c) are licit because in these structures the SUQ takes a bare NP as its complement, but (22b) is illicit because it requires the SUQ to take a NumP as its complement.

In this way, the facts of (16), which are problematic for an NP analysis of NNCs, can be modeled if we allow for a NumP projection in the nominal domain.

6 Conclusion

In this paper, I considered a puzzling aspect of the behavior of English noun-noun coordinate structures. Specifically, these structures pattern syntactically unlike morphosyntactic plurals in that they are compatible with singular universal quantifiers (each and every), but they pattern like plurals for the purposes of subject-verb agreement when they occur with other determiners. I argued that these facts require there to be (at least) a NumP in the nominal projection (contra the NP analysis of nominals).

The analysis sketched in section 5 is not intended as a complete analysis of the structure of the nominal domain. There are many phenomena that I did not touch, numerals, demonstratives, possessives, partitives and pseudopartitives, determiner co-occurrence and phrasal determiners, not to mention the semantics of any of these things. However, there is no reason to think that consideration of these additional factors will force us back into an NP analysis—in fact, there is good reason to think these things are amenable to a DP-analytic account; Heycock & Zamparelli (2005) give an extensive DP-analytic account of numerous of these issues, and an anonymous reviewer suggests Borer’s (2005) account of plural syntax and semantics as another more complete alternative.

Still, one might consider various ways to maintain an NP analysis in light of the arguments presented. One way is to postulate a special number feature for NNCs, call it pl_&, to distinguish NNCs from ordinary plurals. This pl_& feature would behave like ordinary pl for the purposes of subject-verb agreement, allowing us to capture (16a) and (16b), but it would fall together with sg in being compatible with SUQs. We could make a similar move with the denotation of NNCs; maybe they do not have the same denotation as plurals (i.e. as sums), but have some other NNC-specific denotation making them (but not ordinary plurals) compatible with SUQs. However, in the current state of evidence, these moves would be ad hoc and unsatisfying. An anonymous reviewer suggests that, if agreement is semantic (not syntactic) in nature, it might be possible to rehabilitate an NP analysis using the agreement technology of Head-Driven Phrase-Structure Grammar (Pollard & Sag 1994). I leave the question whether such an approach is viable to future work; the conclusions of this paper are therefore restricted to the assumption that agreement does not have semantic effects in English.

Alternatively, we might sever the number-feature facts at issue here from the broader NP/DP debate by adopting an NP analysis where a dependent of N⁰ (not N⁰ itself) projects the number feature for the full NP. This would be fine in principle, but it clashes with the conventional understanding of headedness. At the very least, the evidence presented above requires a departure from a straightforward NP analysis. If section 5 is on the right track, the necessary departure is likely to leave us with an account that is consonant with the independently motivated and highly successful DP analysis.

Acknowledgements

I thank three anonymous reviewers and the associate editor for their careful consideration and constructive comments on this paper. I am also grateful to Alan Ke and Yushi Sugimoto for helpful discussions.

Competing interests

The author has no competing interests to declare.