Semantic agreement, also called sylleptic or ad sensum agreement¹² is found in quantificational or pseudo partitive constructions, or partitive constructions in French and elsewhere. Here, we provide a small sample, but the pattern of agreement we discuss is generally found with proportional quantifiers such as deux tiers, moitié, majorité (two thirds, half, majority), count quantifiers such as deux, douzaine (two, dozen), combining optionally or obligatorily with a determiner, e.g., un, le, ce (a, the, this), followed by de (of) in pseudo-partitives, e.g., groupes d'électeurs (group of voters), or partitives, e.g., (les) deux tiers des étudiants ((the) two thirds of the students), or (une douzaine d'entre nous (a dozen among us), with restrictions as to which combinations of these parts are allowed.¹³

This section briefly summarizes the findings in Charnavel and Sportiche (2025), to which we refer the reader for a more detailed discussion. These findings primarily deal with Subject Verb (Subject Tense) patterns of number and person agreement in the standard SV order in French and English when the subject is a conjunction of two (or more) DPs (with obvious extensions to other cases of DP/head agreement). Here are the conclusions in Charnavel and Sportiche (2025), essentially agreeing with previous authors such as Farkaş and Zec (1995), Doron (2000), or more recently Kučerová (2018) or Harbour (2020).

There are two fundamental reasons motivating these conclusions.

First, agreement is sensitive to the denotational properties of the conjunction, rather than to its syntactic make-up. This is illustrated by the following cases.

Sentences (20-a) and (20-c) are taken to be unsurprising: the subject is a conjunction of two singular DPs, and it is typically assumed that the conjunction is somehow resolved as syntactically plural, triggering plural agreement on the verb. Similarly, in (20-b) and (20-d), the conjunction being resolved as syntactically plural disallows singular agreement on the verb. But some observations–some from Charnavel (2010)–raise challenges for this picture.

Consider the following paradigm of conjoined singular indefinites with un/a or singular definites le, la/ the:

We see that the agreement pattern is insensitive to whether the predicate is collective or distributive. But if we conjoin DPs quantified with chaque / each or aucun / no, the pattern changes as singular agreement becomes acceptable with distributive predicates:

We further observe a contrast between the following cases of conjunction of (in)definites (24) and (25):

The only difference is, again, the denotation of the conjunction of DPs. The denotation of the two conjuncts must be different in (24) since a bachelor cannot be a married man. But they can be the same in (25). And the choice of agreement correlates with the interpretation: with singular, the two conjuncts must have the same denotation. Plural agreement favors the opposite (but does not entirely require it, with an individual seen under different guises).

Second, that agreement with conjunctions is only semantic is corroborated by a diagnostic property of semantic agreement involving existential constructions:²⁰

Example (26-b) illustrates the long-noted (see e.g., Sobin, 1997) ill-formed plural agreement with a conjunction of singulars DPs. This parallels what is observed in example (8-b) with British collective nouns. This contrasts minimally with a case in which agreement can be syntactic in (26-a) which displays the reverse pattern. Such contrasts supports the conclusion that agreement with conjunctions is always semantic.

Another distributional restriction groups together agreement with collective nouns, partitives and pseudo partitive constructions, and conjunctions, namely certain examples of “inverse specificational copular structures”. In a specificational copular structure DP₁ be DP₂ (e.g., The culprit is/*am me, The real problem is/*are your parents), agreement is with DP₁. But an apparent exception to this pattern consists of cases where the initial DP is either a free relative or a relative introduced by all as reported in Heycock (2012, p.212, footnote 3):

Plural agreement is possible in the presence of a plural DP₂ (but not of a singular DP₂), whether indefinite, definite, quantified, etc.. Now, these become deviant with plural agreement when DP₂ is headed by a collective noun, when it is a partitive or pseudo partitive construction, or when it is a conjunction (of singulars):

This pattern can be duplicated in French for the speakers who can omit ce in the sentences in (29). The French analog of the specificational DP₁ be DP₂ normally is DP₁, c'est DP₂ (DP₁, it's DP₂) with either singular default agreement or agreement with DP₂. But in the analog of the examples in (27), ce can be omitted and, with a plural DP₂, singular agreement is disallowed and plural agreement is required:

In such cases, we can reproduce the pattern in (28): with a conjunction of singulars, or a singular (pseudo)-partitive with plural restriction, plural agreement is deviant, and singular agreement is fine.

Interestingly, the three cases we analyze as semantic agreement (collective nouns, (pseudo-)partitives and conjunctions) pattern alike with respect to agreement, and unlike cases of syntactic agreement. How to analyze copular structures and inverse specificational copular structures is highly debated (see e.g., Heycock, 2020), and the exceptional agreement behavior of examples in (27) and (29) remains unanalyzed and beyond the scope of the present article. However, the observed pattern resembles what is found in English existential constructions where syntactic plural agreement is possible with a postverbal DP but semantic agreement is not (cf. examples (26-a) and (26-b)). The observed pattern would thus fall out if only syntactic agreement was possible in examples (27),(28), (29), and (30) but semantic agreement was not. This would fall out if the postverbal element and T are not (cannot be) in a Spec/Head configuration at LF (as some authors propose–see Heycock, 2020).²¹

Agreement satisfies the following descriptive generalizations which should be derived:

We are looking for an economical treatment of these descriptive generalizations, ideally one that requires no new assumptions or stipulations.

There are a number of distinct previous proposals regarding varying agreement with British collective nouns, and a few for (pseudo)-partitive constructions, e.g., Selkirk (1977), Barker (1992), Pollard and Sag (1994), Schwarzschild (1996), den Dikken (2001), Sauerland and Elbourne (2002), Sauerland (2004), Magri (2008), Pearson (2011), Danon (2013), Sportiche (2016), Smith (2017) (similar to Smith, 2021 in relevant respects), Thoms (2019), Messick (2023), and references therein. In section 4, we discuss some of them.

Some proposals are syntax based, other semantics based or a mix.

For the syntax-based proposals, Smith (2017) convincingly critically reviews den Dikken (2001), Sauerland and Elbourne (2002), and Sauerland (2004); and fundamentally based on mismatched simultaneous agreement, Sturt (2022) convincingly critically reviews Pollard and Sag (1994) (and descendants such as Wechsler, 2011), den Dikken (2001), Sauerland and Elbourne (2002) and Smith (2017). So we will limit our discussion in section 4 to Smith (2017) and (Thoms 2019) which represent two different approaches (with some remarks about Magri, 2008, in effect similar to Thoms, 2019).

For the semantics-based proposals, Pearson (2011) reviews aspects of Barker (1992), and Schwarzschild (1996), but none engages with the distributional and interpretive restrictions that are central here, so we will not discuss them.

Of all the previous treatments engaging the restrictions mentioned, all other than Sportiche (2016) are syntactic. In the Agree terminology of Chomsky (2000), a syntactic treatment of such agreements between a T head (a probe) and its DP subject (the goal G) needs the following assumptions:

As far as we can tell, all syntactic approaches face problems with (31-b); in general, they have little to say about generalization (31-c) and none derives (rather than stipulate) generalizations (31-d) and (31-e) except for Thoms (2019). Here, we provide a (non-exhaustive) list with some basic problems they face.

Equidistance approaches (EQAPP) (Sauerland and Elbourne, 2002; Smith, 2017, 2021; Messick, 2023). They can readily deal with mismatched simultaneous agreements, but are too permissive and thus can only handle asymmetries via stipulations for some, or not at all for others.

Structural ambiguities approaches come in two kinds:

Type 1 analyses (SAA1) postulate two completely independent structures, one for each agreement option operating under normal syntactic agreement rules. They go back at least to Selkirk, 1977) and include den Dikken (2001) or Sauerland (2004). They can't deal with asymmetries and mismatched simultaneous agreements.

Type 2 (SAA2) analyses also postulate two distinct structures that are closely related derivationally. They include Magri (2008) and Thoms (2019). They face different problems depending on the version; they can handle the facts in (8) and derive (9) in some (but not all) cases, and deal with some asymmetries, but not all of them; they cannot handle the full range of semantic agreement cases (double movement as in (41)), and overgenerate in some cases.

In addition, none of these approaches deal with semantic agreement with conjunctions. If our conclusion that such agreement is part of a unique paradigm including collective nouns and (pseudo-)partitives, they all fail.

We now present a treatment granting interpretive import features of the probe–more specifically presuppositional properties–recapitulating to some extent conclusions already reached in earlier sections.

Upon hearing out of the blue “she was next to the bank” without knowing anything about the context, one can make a number of inferences regarding what a truthful speaker intends to convey by uttering such a sentence. For example, one can infer that there is a unique relevant bank being talked about, or that the object denoted by she can be spatially located. Concentrating here on the use of the pronoun she, the following inferences can be made:

This is all familiar. We attribute to the pronoun she the features and features values [3^rdperson, -plural, +fem], and these features carry meaning, typically analyzed as presuppositions that, for a well formed sentence, lead to the inferences listed above. More generally, (ignoring gender throughout which raises its own set of issues), let's minimally postulate:²²

A minimal assumption is that this is always true: whenever such features appear in a structure, they always carry these presuppositions. This minimal assumption, the null hypothesis, is widely taken to be false (albeit not in the functionalist literature): ϕ-features on probes are taken to lack interpretive import, a property sometimes seen as driving the obligatoriness of agreement (but not always, see, Deal, 2023 for recent discussion).

Is this departure from minimal assumptions a priori justified? If we think of such features affixed to T, say, as coding a property of T, it is hard to see what kind of interpretive import they could have. But this is not a necessary assumption.²⁴ A verb like sleep presupposes that its subject is capable of sleeping; it restricts the set over which the subject is allowed to range to yield well-formedness. Presuppositional items always restrict one of their arguments so properties on T could restrict the subject of T or the complement of T.

These considerations show that this minimal assumption is not a priori implausible. In full generality, this would mean:

That syntactic and semantic agreements behave differently (e.g., asymmetrically, or with respect to distributional and interpretive properties) probably means that features on T are not always interpreted. In the discussion that follows, we show that a partial version of this null hypothesis, whereby syntactically agreeing features are not interpreted but semantic agreement features are, is predicted under minimal standard assumptions.

We now turn to why semantic agreement is found and discuss why its occurrence is predicted under minimal assumptions.

First, we assume that the Principle of Full Interpretation (FI) (Chomsky, 1986) holds: it requires every symbol present in a syntactic representation at some interface to be interpreted at this interface. Now suppose that a head T is marked with some ϕ-feature values. By FI, if these symbols are present at LF, they must be interpreted. This leads to two possible resolution paths:

So consider the following examples:

In the first example, Agree holds between T and its subject and their ϕ-features match. As a result, the ϕ- features on T are made LF invisible. They have no interpretive import.

In the second example, Agree could hold between T and its subject, but their ϕ-features do not match.²⁶ As a result, the ϕ- features on T cannot be made LF invisible. They must, by FI, have interpretive import, namely the interpretive import they always have (as per (33)). This is semantic agreement as defined in (3).²⁷

The second option, developing a suggestion made in Sportiche (2016) refined in Charnavel and Sportiche (2024), Charnavel et al. (2024), and here, is not a standard assumption, but as we will see below, this assumption derives the properties of semantic agreement. If these features are interpreted, the null hypothesis arises from the minimal assumption we outlined above: they carry their standard presupposition about what they double or enter into an agreement relation with. More specifically:

Consider how this applies to T and its subject DP.

When agreement is semantic, the ϕ-feature values on T and on DP do not have to match, as in the case of British collective nouns, where a singular DP cooccurs with plural agreement. What must be true is that, whatever the syntactic features of the DP are, its interpretation must be compatible with the presupposition carried by the features on T. Importantly, the plural feature value on T does not *make* its Goal DP plurality-denoting. If the DP does not have the internal structure needed to denote a plurality, the result will be ill formed. This will be discussed to some extent in sections 5.1 and 5.2.

When agreement is syntactic agreement, the features of T and of the DP match in value, as is standard. In such a case the features on T can become LF invisible, as is also standard.

Note that, given how semantic agreement functions, the features and their values on T and DP must, in the case of semantic agreement, be generated independently of each other since they can differ. Technical questions arise regarding syntactic agreement, but the simplest implementation (with consequences and potential challenges not discussed here) would take it that with syntactic Agree, feature values on heads/probes and DPs/their goals are generated independently of each other (as in unification based frameworks). This would mean that Agree also always involves feature value matching rather than feature copying or valuing.

If ϕ-features on T are presupposition triggers when agreement is semantic, we expect to see the effects of Heim's 1991 Maximize Presupposition!–henceforth MP. How exactly MP functions and how it should be implemented is unclear; see for example Percus (2006), Sauerland (2008), Singh (2011), and Schlenker (2012)–a debate which lies beyond the scope of the present article. For our present purpose, the following characterization (roughly Singh's) will suffice. Given two sentences S and S' with the same content apart from presupposition triggers, MP requires for each parallel pairs of triggers on the same scale the use of the strongest trigger whose presupposition is satisfied in the context. The use of a weaker trigger implicates that stronger triggers would yield falsehoods.

For Sauerland (2008), which deals with DP interpretation, an interpreted singular specification that requires singular denotation is more informative (stronger) than an interpreted plural specification (which allows both singular and plural denotations). If the singular was true in the context, it would have had to be used. The choice of plural implicates that the (interpreted) singular is false, so the plural ends up being plurality denoting. The behavior of collective nouns or (pseudo-)partitive constructions and agreement raises challenges for how MP functions within such a view. To see why, consider again examples such as:²⁸

In all syntactic treatments, the subject is marked singular in both cases, and there is a plural feature made accessible somehow (but not visibly so in the DP) in (36-b). Note first that uninterpreted singular is less informative than plural since only the latter requires countability. Since is is (or can be) uninterpreted, we might expect that (36-a) cannot mean that the team members are old, since this is more informatively conveyed by (36-b) with plural marking on T. But this is incorrect: sentence (36-a) is ambiguous.

The same problem arises under our account where plural on T is interpreted as a presupposition on the subject. Plural agreement (36-b) requires a plurality denoting subject. Under the proposal in Sauerland, 2008 where plural can mean plurality or singleton denotation, this could be implicated by a singular agreement alternative with otherwise the same content. The relevant alternative is (36-a) with uninterpreted singular on T, hence not more (in fact less) informative than plural agreement in (36-b). So the subject in (36-b) should be able to denote either a plurality or a singleton. This is incorrect, which means we cannot straightforwardly adopt this proposal. We therefore assume here (see, e.g., Harbour, 2014, Mayr, 2015, Harbour, 2016, section 6.4.2, or Mart́ı, 2020 for relevant discussion):

Given (37), interpreted singular and plural do not compete. But interpreted plural and uninterpreted singular do, plural being more informative. Since MP is motivated independently of the theory of number, the ambiguity of (36-a) is now a problem. In it, there is no maximally encoded presuppositional information guaranteeing that the subject denotes a plurality. Since (36-b) is an alternative that guarantees this meaning, the subject of (36-a) should not be able to denote a plurality. But this is incorrect.

In a syntactic approach, the form (36-a) is ambiguous as to whether the plural feature is accessible or not: it can mean the team is old or the team members are old. In (36-b), by contrast, the plural feature must be accessible, triggering plural agreement. To account for the meaning of (36-b), we need to stipulate that accessibility of this plural feature must also mean that the subject must denote the team members only (despite the presence of an interpreted singular as there is only one team) so it means the team members are old: as a result, the two sentences do not have the same relevant content and thus do not compete.

The same fix can be translated in our approach: assume that such a collective noun like team can either mean team or team members. Syntactically singular the team then can either mean the structure or its members.²⁹ This is equivalent to there being an optionally accessible plural feature in the syntactic approach, so that what the form (36-a) conveys is semantically weak: the team or its membership is old. In our approach, the equivalent of assuming that the accessible plural feature in the syntactic approach makes the subject plurality denoting derives from the assumption that T comes with a plural presupposition requiring the subject to be plurality denoting, so that it is not ambiguous (it only means the members are old and therefore (36-b) does not compete with (36-a).

In sum, we expect to find MP effects in cases where an interpreted (presuppositional) plural on T alternates with an uninterpreted singular on T. As we will see, such effects are found.

Generalization (31-a). First, given that syntactic agreement between T and its Goal DP is a purely formal operation where the features on T are not (or need not) be interpreted, it imposes no denotational restrictions on the DP. This means that singular syntactic agreement on T is compatible with the subject being plurality denoting, as long as it is syntactically marked as singular. Semantic agreement, however, does impose such restrictions, since it encodes presuppositions. Plural marking on T does require its subject to be plurality denoting. This derives generalization (31-a) (or (6)).

Generalization (31-b). Consider next how to derive the asymmetric pattern in generalization (31-b).

To account for these data, we capitalize on the difference between syntactic agreement and semantic agreement. Syntactic agreement imposes no restriction on the interpretation of the subject: it may denote a singleton or a plurality, if the internal structure of the subject allows both. Semantic agreement, because it is presuppositional, does imposes denotational restrictions: a DP semantically agreeing with a plural T will have to denote a plurality.

Applying this to (7-a), in example (i), the subject antecedes a plural pronoun, which presupposes an antecedent denoting a plurality. Because the subject can denote a plurality, the result is well-formed. In example (ii), the pronoun presupposes an antecedent denoting a singleton, but the subject must denote a plurality due to the presuppositional plural features on T. This leads to deviance. The same reasoning extends to cases (7-b), and (7-c).

Note that the contrasts in (7a–c) show that anaphors or pronouns bound by the subject can mismatch the agreement on T. This casts doubts on views taking the binding of anaphors (or of pronouns) as involving the same Agree process between anaphor and antecedent as subject/verb agreement. This shows that it is desirable to dissociate syntactic agreement or concord, which is about feature matching, from binding relations, taking the latter to involve denotational covaluation of independent expressions: for a pronoun to be bound, it suffices that its antecedent and the pronoun be covalued; that is, that they denote or range over the same entities (see Sections 5.1 and 5.2 for further discussion).³⁰

Turn now to example (7-d)) (tense twice in conjoined clauses):

This pattern is problematic if it can only be analyzed as involving a single DP subject of a conjoined T', as in (38a) below, as it is unclear why different ordering of the conjuncts would yield the displayed asymmetry.³¹ But another analysis is available as in (38b) and (38c) below whereby the second conjunct has a silent pronoun, which we assume in the non-pro-drop English and French, must be anteceded by the subject of the first one:³²

The contrast now reduces to that seen in (7-a), (7-b), or (7-c). These examples show that a plural agreeing collective noun headed DP cannot antecede (or bind) a singular pronoun, it can only antecede a plural pronoun. And a plural pronoun can only trigger plural agreement, and a singular pronoun can only trigger singular agreement, even when anteceded or bound by a collective noun headed DP that can trigger both agreement. This can be illustrated as:

Now in (38), [the group]_k can denote a plurality so e_k can be plural and antecede a (silent) plural pronoun triggering plural agreement (or a singular pronoun triggering singular agreement). Given the plural features on T, [the group]_m must denote a plurality. As such, it can only antecede a plural pronoun. But a plural pronoun cannot trigger singular agreement. Since [the group]_m cannot antecede a (silent) singular pronoun, singular agreement on T, (7-d)b is ruled out.

Let us now turn to examples (7-e) (tense twice (relative/main clause)):

If the relativized subject agrees in the singular as in (7-e)(i), this is syntactic agreement which imposes no restriction on the denotation of this subject DP. As a result, there is no DP internal restriction imposing any denotational restriction on the whole relative clause. In particular, it can denote a plurality, and in fact it must, given the plural agreement in the main clause. Fundamentally, this example behaves like the committee are meeting.

By contrast, in (7-e)(ii), plural agreement requires the relativized subject to denote a plurality. The whole relative clause must denote a plurality too. In such a case, singular agreement is excluded. But this is not derived. Something additional is needed.

Examples (7-e) form a minimal pair with examples (36) repeated below:

In (36), there is no property internal to the DP team that makes it unambiguous and both agreements are possible. Similarly, in (7-e)(i), there is nothing internal to the DP subject that makes it unambiguous since relative clause internal agreement is singular and can therefore be syntactic; and both singular and plural agreement in the main clause are possible. But in (7-e)(ii), relative clause internal agreement is plural, semantically requiring committee to denote a plurality, independently of main clause agreement. As a result, in the following pair, the subject must be unambiguously plural denoting (unlike in (36)) and the two sentences without the plural presupposition expressed by the main T have the same relevant content:

Since main plural agreement triggers a stronger presupposition than (uninterpreted) singular agreement, Maximize Presupposition, as discussed in Section 3.3 is relevant: it requires the use of the stronger trigger, here plural, excluding (7-e)(ii) (= (40-b)).³³

Turn next to example (7-f) (tense and reflexives in conjunction):

Again, a single DP subject of a T' conjunction as sole analysis is not possible since plural agreement would restrict it to a plurality denoting DP, incompatible with the singular reflexive. Instead, we assume that the second conjunct can have a silent pronominal subject, marked e here, anteceded in the first conjunct.

To understand this case we asked first how (α) the government defended itself from the scandal is interpreted. This can mean the institution or its members defended the institution or themselves. As should be clear, the singular feature on government is interpreted–there is only one government being talked about (see further discussion in Section 4.1.1)–but this does not mean that the denotation of the subject is a singleton. Similarly, the singular marked itself does not force a singular denotation on the subject or on the object, they can both denote a plurality too due to the ability of these collective nouns to come with a plurality presupposition. Expanding the structure of the pronoun it as a definite description with a silent NP (cf. Elbourne, 2001); this is analyzed³⁴ as the government defended [the government]self with government denoting either the institution or its members. Turn now to (7-f-i). The first conjunct (reportedly) has the same meaning as (α). As a result, the silent pronoun can choose to pick out the plurality denoting option as value.

By contrast in (7-f-ii), due to the plural reflexive, a plurality denotation presupposition must be present on the subject of the first conjunct mandating a plural denotation only. It cannot mean that the institution defended its members from the scandal. This subject cannot antecede a singular pronoun (cf. **(7-a)ii) resulting in ill formedness.

The last examples (7-g) involving relative clauses, can be similarly analyzed:

This discussion straightforwardly extends to the parallel cases of partitive and pseudo partitives constructions, for example, une majorité de (/des) maisons / a majority of (the) houses: even if their head (here majorité/majority) is syntactically singular, they are compatible with plural agreement if interpreted as denoting a plurality.

Finally, this account makes a correct prediction for French, which, due to its richer morphological exponence of agreement, illustrates a restriction not visible in English. Uniform agreement with two distinct heads is sometimes required (compare with (7-g) or (7-d)).

Here, the two heads must agree in ϕ-feature number values: both singular or both plural. Why? Agreement with the two heads is with a single moving DP. Plural marking on any head requires this DP to be interpreted as plural denoting. Now MP requires the use of the strongest trigger whose presupposition is satisfied in the context. Uninterpreted singular agreement on the other head would not maximally encode this presupposition. So plural on the other head is required too (this also applies to the case in footnote 31).

Generalization (31-c). This generalization follows from the hypothesis there is no syntactic feature resolution in conjunctions that yield ϕ-feature values for the whole conjunction. As a result, agreement with a conjunction is necessarily semantic. A conjoined subject has to satisfy the presuppositional properties of the ϕ-feature values appearing on T.

Generalizations (31-d) and (31-e). The latter, (31-e), follows from the hypothesis that the features on the probe are presuppositional in cases of semantic agreement: since presuppositions are interpretive properties they must be satisfied at LF. The former follows from general properties of presupposition satisfaction. Indeed, a presupposition trigger is a function restricting the domain over which (one of) its argument can range. That is, a presupposition trigger imposes restrictions on its sister or its subject at LF. Fundamentally, this is a subcase of the general Locality of Selection Principle or the strict locality of function/argument relations. Now T takes two arguments: a DP subject as syntactic argument and a non DP (say a VP) as complement as syntactic and semantic argument. If ϕ-features on T are presuppositional and given the nature of T and of the ϕ-features on T, the only option for these ϕ-features is to constrain the denotational possibilities of the subject of T at LF. If there is no subject e.g., due to reconstruction, the trigger is a function without argument, which is ill formed. This derives generalization (31-d).

Consider DPs headed by a collective noun such as a team that tolerate more than one agreement option.

First, recall that in all such cases, the nouns like team, ... are singular, only tolerate a singular determiner, and are interpreted as referencing a single token: a single team. This shows that the minimal DPs containing them are both morphologically and semantically singular: we cannot be dealing with a case of number ambiguity, underspecification, or uninterpreted number property for this singular noun.

The noun team, also allows its immediate DP container to denote not (just) the singular team entity but the plurality of its members.³⁵ This plural property is a lexical property of such collective nouns distinguishing them from other collective nouns (see Levin, 2001, for documented variation in this respect in several English varieties) or nouns such as, e.g., canopy, which reference the layer composed of the uppermost branches of the trees in a forest, a discrete set, but cannot trigger plural agreement. This lexical property also distinguishes such nouns from their cognate in other languages, e.g., French: French une équipe / a team also denotes an entity composed of members, with reference to these members possible (cf., l'équipe n'a pas voté de façon unanime / the team did not vote unanimously, where the membership is interpretively referenced) but such nouns, lacking this lexical property, do not allow plural agreement. In a syntactic approach, this property can be viewed as team being marked with a feature, ϕpluralm, call it mereological to adopt the terminology of Elbourne (1999), that encodes this possible plurality: this references the property of teams to be composed of more than one part, their members (recall that we code this property as a presupposition). What structure might such DPs containing a collective noun be like? Consider a DP such as this team: where is the singular/plural property of team/teams encoded? The standard view encodes it in a NumberP projection taking the NP team as complement. This feature has both morphophonological reflexes (e.g., the plural -s suffix), and semantic reflexes, plural for example meaning that the DP denotes a plurality of teams.

This yields the following kind of structure for the subjects of (2), i.e. for a singular (assuming a, this, belong to the category Determiner) where the determiners must agree in ϕ-feature values with Num, and hence are singular here:³⁶

Superficially, in a French or English pseudo-partitive or partitive construction, such as a majority of (the) girls, agreement can be either with the quantity noun majority or with it domain of quantification (the) girls. Agreement with the latter is exclusively semantic, displaying different properties than agreement with the former, which is, or can be syntactic.

The structure of quantificational/ pseudo partitive/partitive cases is substantially debated (see, e.g., Falco and Zamparelli, 2019, and references therein) but the details of the debates mostly do not impact what is relevant here. Nouns like majorité, moitié, ../ majority, half in French are lexically unmarked for number and can appear in the singular or the plural coded in their NumP, with this number property phonologically and semantically interpreted.

A pseudo-partitive case like une majorité de filles/ a majority of girls, which allow singular or plural agreement, can, a priori, be analyzed as one of the two options (49a,b) (assuming une, des/ a, indefinite plural, belong to the category Determiner), one right branching only, the other not:

A partitive part/whole structure is roughly assumed to be built on a pseudo partitive structure: a majority of the girls is treated in effect as a majority of N of the girls, or two of the girls as two N of the girls, where the deleted head may be the same noun as that denoting the whole (here books) or something less specified compatible with its denotation.⁴⁶ This yields two options for partitives like une majorité des filles/ a majority of the girls:

For (pseudo-) partitives with numerals like two of the books, we can, for our purposes here assume a simpler structure as what matters is that only the singular or plural Number property of the overall DP matters and is the only one syntactically accessible (but cf. Danon, 2012 for discussion):

Postulating a structural ambiguity in connection with agreement and other facts goes back at least to Selkirk (1977). Selkirk's 1977 proposal is based on the observations in (a, b, c):

The reasoning is that selectional restrictions in (52c) indicate that the head of a bottle of wine is bottle with break, vs. wine with drink, with this structural ambiguity taken to feed the agreement differences in (52a) and (52b). But note that (52d) is fine, with bouteille de vin a shared constituent, regardless of how this RNR structure is analyzed (be it a truly shared constituent, or backwards ellipsis, as in Barros and Vicente, 2011). Example (52e) makes the same point. Example (52f) where the noun bottle appears to be modified by an adjective modifying its size, and another qualifying its content shows that a structural ambiguity approach is at best unnecessary, as Danon (2013) concludes. The facts in (52b) reproduce the behavior of of pronominal binding vs verbal agreement and are not telling.⁴⁷

Now if both the (52a) and (52b) structures above are available or the second member of each pair is, the quantity noun and the domain constituent would be equally accessible to an outside probe, consistent with either structural ambiguity approaches or equidistance approaches. But this is an undesirable result since it would predict identical behavior for the two agreement possibilities. This means that only structures (52a) should be available. General phrase structure theoretic principles a priori favor the (52a) member of each pair, as in (52b) member, the overall DP is not headed (as surely X is not a D), and some additional mechanism must be provided to explain why such structures are, say, definite or indefinite depending only on the choice of an embedded D or why standard syntactic agreement is always consistent with this choice of D. Furthermore, there is crosslinguistic evidence supporting the idea that only the (52a) structural analysis is available, discussed in Danon (2013) which concludes (specifically for Hebrew, but surveying a variety of languages) that there is no such structural ambiguity.

Note finally examples such as (52g) where the noun majorité/majority is simultaneously modified by large, relating to the its quantity meaning, and by silent, related to a property of its members: this indicates that a structural ambiguity based on the quantifier being either purely quantifying or referencing an abstract entity of a certain size, would be insufficient.

We conclude that a single structure underlies these (pseudo-) partitive constructions, namely the (52a) member of the pairs above.

This means that structural ambiguity approaches of type 1 (SAA1) based on different hierarchical organizations for each agreement option is not available.

How about equidistance approaches? To treat varying agreement, Danon (2013) postulates an upward feature percolation mechanism whereby the low plural property can percolate up to become available for DP external agreement (so in (50-a) for example, ϕ_pl in X can percolate up to Num of majority). This is an SAA1 approach based not on different hierarchies but on different content. Danon (2013) does not address mismatched simultaneous agreement, but Messick (2023), who adopts this percolation mechanism and reports mismatched simultaneous agreement for English,⁴⁸ does. Messick's treatment is quite similar to Smith's, using two sets of features (as in Wechsler and Zlatić, 2003) but with some differences: it assumes that anaphor–antecedent agreement involves feature agreement just like in Smith's proposal, but such agreement is countercyclic: it applies after T agrees with its subject. In addition, it stipulates a Constraint on Agreement with Semantic Features (CASF) stating that ‘once semantic features have been accessed for an agreement operation, all other subsequent agreement operations must target the semantic features'. The conjunction of these assumptions derives some of the data in (7), e.g., (7-a), (7-b) or (7-c). For example, the committee are proud of itself is out because plural agreement on T is semantic, and by CASF, subsequent agreement between the subject and the anaphor must be semantic too.

But such a treatment fails on a number of points. First, it cannot handle any of cross sentential asymmetries which only involve two T agreements without derivational connections, e.g., (7-d), (7-f) (and perhaps also (7-e) and (7-g) although this depends on further assumptions not discussed in Messick, 2023). It is unclear what exactly the distinction between syntactic and semantic features is, since, as repeatedly stressed, the syntactic number features are interpreted. It is also unclear why CASF should hold at all, or why this constraint applies to semantic features rather to syntactic features. It requires countercyclicity for antecedent/anaphor agreement. And it is unclear how countercyclic agreement applies in the case of bound pronouns, which can be indefinitely distant from their antecedent, (cf. e.g., *No team_k are proud of the results (the news report) it_k got) can be treated unless Agree is an unbounded relation.⁴⁹ It cannot handle the French data in (41): an example like La majorité des électeurs sont loyale / the majority of the voters are loyal-sg should be OK since low agreement with the adjective would be syntactic, high agreement semantic, consistent with CASF. And, since in the end, semantic agreement is treated as syntactic agreement with semantic features, it says nothing about distributional and interpretive restrictions in (16) or (30-b).

Finally, the Magri-Thoms approach to collective nouns does generalize straightforwardly to (pseudo-)partitives: the sortal NP voters is merged as argument of voted, and a majority of is added later, appearing on the spine only above T.

Consider now (12) and (13), repeated below:

In these examples, what is agreement with? Could it be with the DP denoting the whole? If this was the case, we could explain why we get first or second person agreement in the presence of nous, vous /us, you-pl. To explain the possibility of third person agreement, we would have to assume that nous/us is actually composed of a first person head above a plural pronoun: nous/us = [ 1^stpers [ eux/them]], not an implausible assumption. Agreement in the third person plural would come about if only eux/them was merged lower that T and everything else in the subject DP above T in the spine.

This approach, however, makes incorrect predictions.

First, if plural agreement is only (somehow) syntactically linked to the features of the pronouns nous, vous, eux / us, you-pl, them, this predicts that what occurs above these pronouns in the DP is irrelevant. But this prediction is false. Thus, there is a minimal contrast between the following two sentences. Consider:

In (53a), the subject DP denotes a singleton due to the presence of the numeral/indefinite article un and plural agreement is excluded. In (53b), with the numeral deux/two, plural agreement is allowed. The cardinality of the denotation of the whole subject matters. Even worse, there are cases in which the quantity restriction is, in principle, compatible with plural agreement, but plural agreement fails if the DP denotes a singleton:⁵⁰

Here, the sortal NP is plural, and one half normally allows plural agreement with plural sortal NPs, but the overall denotation is singular because it is one half of two. This falls out under a semantic agreement approach, but seems out of reach of syntactic approaches.

Secondly, under such a derivation, the argument of the verbal predicate is the DP denoting the whole, not the DP denoting the relevant part of the whole. This yields incorrect results: it would treat (i) a majority of us are sick as meaning (ii) a majority of us is us who are sick that is, a majority of a group including the speaker is a group including the speaker who are sick. However, (i) can be true and (ii) false if the sick majority of us does not include the speaker. But such an agreement cannot be prevented under Thoms's proposal: it suffices to merge nous low, and everything else (namely, une majorité d'entre) above T. So this proposal overgenerates.

One conclusion emerging from this discussion of person agreement is the following: if agreement is syntactically with anything, it can only be with the constituent denoting the sortal NP, namely, what codes the domain of quantification marked NP_domain in structures (49-a), (49-b), (50-a), (50-b). We discuss how person agreement can be handled in Section 5.1.

Overall, the Magri–Thoms structural ambiguity treatments are substantially more successful than the equidistance approaches or the other structural ambiguity approaches. But they face serious difficulties (and, in Thoms's version, requires the availability of External Merge which overgenerates): namely, they require stipulating (6), they cannot handle the agreement data in more complex cases such as some mismatched simultaneous cases, or agreement with two heads under movement as (41), and they cannot handle person agreement in French partitive constructions. If generalization (6), or whatever derives it, is all that is needed, as we argue in Section 3, the structural ambiguity they postulate, together with the added assumptions to make it possible, is unnecessary.

At the end of section 4.2, we reached the following conclusion: in examples (49-a), (49-b), (50-a), (50-b), if agreement could syntactically be with anything, it could only be with the constituent denoting the sortal NP, namely what codes the domain of quantification marked NP_domain in structures (49-a), (49-b), (50-a), (50-b).

Our analysis is that agreement is semantic since this NP is too deeply embedded to be accessible to syntactic agreement. But how is the right denotation delivered in a way consistent with the presupposition of the ϕ-feature values on T, e.g., when it carries first or second person values? More specifically, consider again an example such as

Recall that the different agreements yield different meanings: sommes requires the half to include the speaker, sont requires the half to be larger than one (so is out with nous deux/us two), and est is neutral. How exactly does the well-formedness of (55) with 1pl or 3pl agreement come about? First, we analyze nous (us) as person marked pronoun (cf. Postal, 1969, Elbourne, 2001): it includes the definite article, Number, first person (1p), and a silent NP. When T is 1p, either DP* is 1p, in which case the presupposition on T is trivially satisfied (as with we), or DP* is compatible with 1p: in this case, it denotes a set of individuals including the speaker. But if the latter, it should make no difference if nous six (us six) is replaced, e.g., by the men as long as the speaker is a man (the 1pl presupposition would be met since the set of men includes the speaker). However, the result is deviant:

What then allows (plural and) person agreement in (55)? The presence of nous (us) is crucial but it can't be just nous (us): in cases such as un de nous deux (one of us), neither plural not person agreement is allowed (see (58-a)); similarly in une moitié de nous deux (one half of us two). To tackle this question, we have to take into account the syntactic analysis of partitives. Now this partitive subject is analyzed as shown in (50):

where the silent X* is anaphoric on X**, a subconstituent of nous /us. Since the overall DP is third person singular, it must be X* that encodes 1p and pl. This means, as argued in Harbour (2014), Mayr (2015), Harbour (2016) and Ackema and Neeleman (2018) but not in Déchaine and Wiltschko (2002), person is coded at the NP level (within X), not at the DP level. And in the present case, the denotation of DP* is consistent with the presupposition imposed by T.

Now une moitié de nous deux sont (one half of us two are...) is correctly excluded because the plural presupposition on NP* conflicts with the overall denotation of DP* (which is a singleton). But why is the following example excluded?

Here, DP* would denote a single individual: since X* could be 1sg, 1sg agreement on T should be well formed. This means that there must be a restriction preventing X* from being 1sg here. If X** is structured as [1person [ plural NP]] (as in Harbour, 2016), it does not contain a constituent X*=[1person [singular NP]] could be anaphoric on.⁵¹

It may also be related to a more general phenomenon, also illustrated by the ill-formedness of (56), namely the cases discussed in the next section of definite descriptions (imposters) picking out the speaker, but incompatible with person agreement on T (see footnote 55) or prohibited from binding first person pronouns.

In sum, to handle agreement in the cases discussed in this section, we conclude in agreement with Harbour, 2014, Mayr, 2015, Harbour (2016) and Ackema and Neeleman (2018) that the person property of pronominal DP is not a high property within these DPs but rather a low property of NPs. This means that person presuppositions should be reformulated as applying to NPs.

A reviewer (rightly) wonders why the following kind of examples, some involving imposters or camouflage in the sense of Collins and Postal (2012), are consistently ill-formed with first person agreement:

In this section, we consider this question and some connected issues, briefly outlining why we think these facts (and some others) are compatible with our general approach. Needless to say, this brief discussion cannot do justice to the relevant array of intricate data connecting the internal syntactic structure of such expressions (cross-linguistically), the semantic analysis of person features, how they are interpreted, and how their interpretation connects in detail with agreement and binding, nor to the proposals in Podobryaev (2017) addressing some of these questions. Some of this is discussed in more detail in Charnavel et al. (2025).

Note first that it is not clear whether the restriction in (59) is a general property or whether it is subject to language variation and, if so, what this variation correlates with. The relevant literature discusses imposters or so called “unagreement” configurations where, for example, subject definite descriptions *seem* compatible with first person agreement and are understood to include the speaker in their denotations (see, e.g., Collins, 2014, Höhn, 2016, Ackema and Neeleman, 2018), but there is no consensus on how to analyze these data (in particular syntactically).

Limiting ourselves to English (as standardly reported) and French (as we have found it from our survey of speakers), let us grant the standard answer that the subjects in (59) are not (and perhaps cannot be) morphosyntactically first person (therefore cannot trigger first person agreement under a syntactic approach). In our approach, this translates to these imposters (or definite descriptions such as in (56)) not being compatible with the first person presupposition imposed by T. If there turned out not to be real variation, we would want to explain why such elements cannot be marked first person or meet a first person presupposition. Under either approach, since first person is interpreted, this suggests that first person interpretation/presupposition is stronger than requiring reference to the speaker (see footnote 56 for some speculative remarks).

But this issue is complicated by (at least) two factors (and many more ignored here, given the reported cross-linguistic variation, e.g., in Collins, 2014).

First, some imposters can trigger person agreement. Recall the plural of “politeness” in French where a single addressee can be addressed by a form identical to the second person plural pronoun with concomitant agreement: Vous êtes là (you-2pl-polite are-2pl there).⁵² Now, French expressions of deference such as Votre Altesse (Highness), Éminence (Eminence), Excellence (Excellency), Grâce (Grace), Majesté (Majesty), Seigneurie (Lord) allow both third person agreement and polite agreement:

Prescription requires third person agreement, but many speakers (all we consulted) allow polite agreement here⁵³ contrasting with the French equivalents of examples in (59) which are sharply out with first person agreement. The existence of such cases underscores the fact that there is no organic prohibition of semantic agreement with imposters.

The behavior of these imposters can be captured by assuming that such expressions are morphosyntactically singular but can have a second person presupposition. This correctly predicts the usual pattern of mismatched simultaneous agreement seen earlier. In the examples below, matching agreement between T and the pronoun (her/your) is preferred, but mismatches follow the pattern we have already encountered (with number): replacing ses / her by vos / your is relatively acceptable in (61-a), but the other switch in (61-b) is unacceptable.⁵⁴

Naturally, it would be desirable to understand whether this difference between first and second person, be it handled syntactically or in our terms, is accidental, and if not, why it holds. More cross-linguistic research is needed to answer this question.

Second, there is at least some asymmetry between singular and plural exemplified here with first person singular and first person plural binding possibilities documented in Collins and Postal (2012) and in Podobryaev (2017) (and is also reported for some languages in Collins, 2014) and illustrated below–a child is addressed by her father for the first two):

Covaluation is out in the first example but it is fine in the second (where binding negates the alternatives: nobody else thinks they should get ready) and the third. And this asymmetry extends to agreement (some of this is noted in Collins and Postal, 2012):

This cannot be due to a difference between the singular your humble servant and its plural given the contrast between (63-a) and (63-c). This second observation suggests an asymmetry at least between I and we and between suis and sommes. Given that we require matching presuppositions between binder and bindee or between semantically agreeing T and DP, this means that first person plural presupposition is more tolerant, weaker, than first person singular presupposition. There are various ways of getting this result, although it is premature given that the full crosslinguistic empirical picture (how general the facts in (59) are) is unknown.⁵⁵

We adopt a presuppositional approach to the binding and agreement patterns. Collins and Postal (2012) argue that the optionality found in (i) Mommy and Daddy will enjoy ourselves/themselves in the Bahamas challenge presuppositional approaches accepting MP: the first and the third person pronouns in (i) should be in competition, and the first person pronoun should win since it maximizes presupposition. We think such examples are in fact not problematic (and neither does Podobryaev, 2017, p. 344, fn 14, but for a different reason) because we take it that point of view/perspective/logophoricity plays a crucial role in binding relations (cf., Sportiche, 2022): the subject Mommy and Daddy can either be presented from the point of view of the child (the way you call me), in which case the pronoun must too (and thus be themselves, at least in a language without indexical shift); or it can be presented from the point of the speaker (a father addressing a child: the way I call myself when speaking to you) in which case the pronoun must too (and thus be ourselves).⁵⁶

In sum, we adopt a presuppositional approach to the binding and agreement patterns that needs to be refined as compared to some standard approaches but is able to handle the relevant facts.

Many questions remain and, as noted, these remarks do not do justice to the full range of data discussed in Collins and Postal (2012) including their “homogeneity” condition or the version of this condition defended in Podobryaev (2017). We again take it that this condition is due to constraints on perspective shifts (the facts are highly reminiscent of the distribution of logophoric pronouns in some African languages), but this discussion falls outside of the scope of the present article (but cf. Charnavel et al., 2025).

As final note, the special status of person as compared to number is highlighted by some additional observations suggesting that person agreement is in fact always semantic (which, if true, would mean that Agree is blind to person features or can't make person features LF invisible) as it satisfies generalizations (31-d) and (31-e), that is, There is always me /* There am always me, with excluded first person agreement. Similarly, paralleling examples in (27), (28), (29), and (30), but with person agreement, we find: (i) Tout que je peux voir est/*êtes Votre Majesté (All I could see is/*are-2pl Your Majesty).⁵⁷ This conclusion is consistent with the descriptive generalizations concerning person agreement in Baker; Baker's 2008 or 2011 S(tructural) C(ondition) O(n) P(erson) A(greement), or den Dikken (2019)'s (for whom it always requires a Spec/Head licensing configuration). This is also consistent with neural evidence (from ERP and fMRI experiments) showing that violation of person but not (standard) number agreement induces e.g., N400 effects usually associated with interpretative problems (Mancini et al., 2011, Mancini et al., 2017).