Two Dutch many’s and the structure of pseudo-partitives

This article discusses the syntax and semantics of Dutch pseudo-partitive constructions with measure nouns, such as drie liter water ‘three liters of water’. The major empirical puzzle is the distribution of two Dutch instances of many/much: veel and vele. Unlike earlier proposals, I analyze veel as a gradable adjective, and vele as a numeral. It turns out that in pseudo-partitives with pure measure readings, only vele can freely occur (veel liters water only allows a marked “liter-bottle” reading). This is puzzling, because veel is otherwise allowed both with mass and count terms, and both in the singular and in the plural. I adopt the more-or-less standard right-branching syntax for Dutch pseudo-partitives (providing some new arguments for its correctness), and propose a semantics for measure nouns which, in combination with Ionin & Matushansky’s semantics for cardinals, correctly characterizes these constructions and explains where pure measure readings occur. I then show that my analysis correctly derives the behavior of veel and vele in these constructions, given their characterization as a gradable adjective and a numeral.


Introduction
Dutch has two instances of many/much: inflected vele and exceptionally uninflected veel, which differ in several respects, both syntactically and semantically. Central to the present article is the novel observation that only vele can normally combine with measure nouns in pseudo-partitive constructions. Consider the data in (1) and (2): (1) a. veel boeken veel many books b. veel wijn veel much wine c. #veel liters wijn veel many liters wine 'many one-liter-units of wine' (2) a. vele boeken vele many books b. *vele wijn vele many wine c. vele liters wijn vele many liters wine 'many liters of wine' The uninflected variant veel in (1) can combine both with count nouns and with mass nouns, but in combination with a measure noun in the pseudo-partitive construction (1c) it does not allow a normal measure reading. (1c) cannot refer to many liters of wine, but only to many discrete one-liter units (e.g., bottles) of wine. The inflected variant vele, on the other hand, which can combine with count nouns but not with mass nouns, does allow a normal measure reading for (2c).
The goal of this article is to argue for a particular syntactic and semantic analysis of Dutch pseudo-partitive constructions which, in combination with the semantic and syntactic characterization of veel and vele that I will propose, yields an explanation for the pattern illustrated in (1) and (2). Although some of the evidence for my analysis of veel and vele depends on data involving pseudo-partitives, and vice versa, I will try to separate out the issues as follows. In section 2 I describe the major differences between veel and vele. I discuss their distribution, and conclude that veel is most plausibly characterized as an adjective, whereas vele patterns like a vague numeral. I also observe some semantic differences between the two elements, demonstrating that veel is gradable, but vele is not. In section 3, I turn to the Dutch pseudo-partitive construction. I adopt the standard rightbranching analysis of Dutch pseudo-partitives, and provide some novel evidence in its favor. I then argue for a corresponding semantic analysis of pseudo-partitives which, in section 3.3, I combine with the semantic characterization of veel and vele provided earlier to arrive at an explanation of the data in (1) and (2). Section 4 briefly discusses how veel and vele behave with respect to the proportional/cardinal distinction, comparing them with the multiple instances of many in Russian and English discussed elsewhere. 1 2 Two Dutch many's

Two Dutch many's and prenominal inflection
Inflection on Dutch prenominal elements is determined by number, gender, and definiteness. Every slot in the paradigm receives an -e (schwa) ending, except for the singular neuter indefinite case, where the ending is -Ø, as shown in table 1. 2 (i) Ik ben veel in Amsterdam geweest. I am much in Amsterdam been 'I have visited Amsterdam a lot' (ii) Ik heb met veel rekening gehouden. I have with much account held 'I have taken much into account' 2 The situation in table 1 is sometimes described as involving strong inflection (showing distinctions for gender and number) in the indefinite, versus weak inflection (showing the -e ending everywhere) in the definite. The same patttern is found in other Germanic languages; e.g., Swedish and Danish also have strong inflection on adjectives (showing three different endings for number and gender) in the indefinite, and weak inflection in the definite; see Schoorlemmer (2009) for recent discussion. The situation in German is rather different, in that both definite and indefinite contexts can lead to stronger and weaker inflection on the adjective, depending on the inflectional richness visible on the preceding determiner.  Ruys: Two Dutch many's and the structure of pseudo-partitives Art. 7, page 3 of 33 Booij (1992) describes these facts with lexical insertion rules that spell out [sg,Nt,indef] as -Ø, with -e the elsewhere case. Schoorlemmer (2009) also has -e as the elsewhere case, and [sg,Nt] spelled out as -Ø; the definite determiner blocks DP-internal agreement so also gives the -e. Menuzzi (1994) and Kester (1996) take the opposite approach: plural number, common gender, and definiteness each specify the presence of the -e, and -Ø appears when all three are absent.
For prenominal elements that inflect, inflection is not optional. However, there are two variants of veel ('many/much'), one of which shows inflection and one of which does not (a similar pattern is found with weinig, 'few/little') (I add an adjective to remind the reader of the expected inflection): 3 shows that one also appears in the indefinite plural, alongside the expected form with an -e ending in (3b). An uninflected form also unexpectedly appears in the singular common indefinite, as shown in (3c) (the singular in fact blocks the inflected variant, as discussed below). There is no obvious semantic difference between (3a) and (3b), although (3b) is often described as being more formal. As will become clear, however, we are not dealing with a single lexical item whose inflection is optional or conditioned by the local syntactic context. We will see instead that there are two lexical items with significant semantic differences. For instance, as discussed in section 2.3 below, the uninflected variant is gradable, the inflected variant is not. Also, while both elements allow a cardinal reading, only the uninflected one naturally allows a proportional reading (see section 4). My discussion of the distribution of the two variants will lead to the conclusion that the uninflected variant of (3a) behaves more as an adjective, and the inflected variant of (3b) as a numeral; for this reason, I will label them as veel A and vele Nl , respectively. I will retain this notation even when arguing, in section 2.2 below, that there is one context where veel A does inflect.
The existing literature observes a few additional distinguishing properties. Uninflected veel A cannot be preceded by a definite determiner or a possessive: (4) a. de *veel / vele mooi-e boeken the many nice books b. Jans *veel / vele ernstig-e tekortkomingen John's many serious shortcomings Kester (1996: 107) suggests that the uninflected form veel, which cannot be preceded by a determiner, is a quantifier, whereas the inflected form, which can, has adjectival status (see also Broekhuis 2013: 283). The assumption that inflected vele Nl is an adjective would indeed explain why it can be preceded by a determiner in (4), and why it must bear adjectival inflection. And the assumption that veel A is syntactically a quantifier with a distribution similar to that of elke 'every', etc., will explain why it cannot be preceded by a determiner in (4). However, as Kester admits, taking uninflected veel A to be a quantifier does not explain why it does not inflect, since other quantifiers (elk(e) 'every', ieder(e) 'every') do.
Kester also postulates a semantic distinction: veel A allows a collective reading, but vele Nl is always distributive. This serves to explain why only veel A can combine with mass nouns, as (5a) shows: 4 (5) a. veel A / *vele Nl lekker-e wijn sg,C much nice wine b. veel A lekker-Ø bier sg,Nt much nice beer It bears repeating, however, that the distinction between veel A and vele Nl goes beyond the mass/count distinction observed with English much/many: veel A occurs both with count and with mass terms (but not with measure nouns), and there are further distinctions, notably with regard to gradability, as we will see. Note, incidentally, that given the incompatibility of vele Nl with mass nouns, the form veel in (5b) cannot be analyzed as an instance of vele Nl bearing the -Ø inflection triggered by the singular neuter indefinite environment; it can only be analyzed as veel A . Being incompatible with mass nouns, vele Nl is banned from singular environments, so it never bears the -Ø inflection.
To summarize, the picture that emerges from the literature (also Haeseryn et al. 1997) is that uninflected veel A is a quantifier of the every-category higher up in the DP, in complementary distribution with determiners and other quantifiers, and inflected vele Nl is a distributive adjective lower in the DP. In the next section I will show that this description does not cover certain exceptions to the pattern in (5). I will argue that virtually the opposite theory is to be preferred: veel A is most likely a gradable adjective; vele Nl is more akin to a numeral (hence the labels). Section 2.3 provides evidence for the gradability of veel A .

Inflected veel A with mass nouns
While (5a) shows that inflected vele Nl is blocked with mass nouns in indefinites, we seem to observe the opposite pattern in definite DPs with mass nouns, where we do find an inflected form:  Kester (1996: 108) also reports that veel A allows a collective reading in (i), but vele Nl does not; Broekhuis (2013: 284) reports that the intuition is shared by "many speakers":  (6) can be neither "quantificational" veel A , which supposedly does not inflect and does not cooccur with determiners, nor "adjectival" vele Nl , which does not combine with mass nouns. There is no obvious way out: it is difficult to understand how the semantic incompatibility of vele Nl with mass nouns could be overcome by making the DP definite, or how the complementary distribution between veel A and the definite determiner could be abrogated in the context of a mass noun. I propose a virtual reversal of the relative positions of veel A and vele Nl in the DP. Inflected vele Nl is a vague numeral, on a par with meerdere 'several', enkele 'some', ettelijke 'many', luttele 'few', verschillende 'various', which also inflect. 5 Uninflected veel A is not a quantifier or a determiner, but -presumably -an adjective (or possibly also a numeral). This will allow us to capture the data observed so far along the following lines.
To explain how the inflected form vele, which did not combine with a mass term in (5a), can combine with a mass term in (6), I diagnose vele in (6) as an instance of veel A , not vele Nl (for independent evidence, see the discussion surrounding (23) in section 2.3 below). This diagnosis is possible, since we are now assuming that veel A is not a determiner or a quantifier but an adjective: this allows it to cooccur with the determiner in (6). Diagnosing vele in (6) as an instance of veel A requires, of course that we assume that veel A is not fully uninflected: it shows the -e ending, but exclusively for the feature [+definite]. This explains why it appeared to be in complementary distribution with the definite determiner and the possessive in (4): veel A does appear in (4), but receives the -e ending and becomes indistinguishable from vele Nl . In this manner, we explain why the inflected form can appear with mass terms, but only in the definite: veel A combines with mass terms, and appears as veel in the indefinite (5a) and as vele in the definite (6). 6 If veel A is to be an adjective, we must accept that not all adjectives show the full inflectional paradigm (but recall that the earlier assumption that veel A was a determiner or quantifier also did not explain why it does not inflect). This is not at all uncommon, however. Booij (1992) and Odijk (1992) discuss several classes of adjectives with an incomplete inflectional paradigm. Some adjectives never inflect, sometimes for phonological reasons. In other cases, the presence or absence of the -e ending reflects a semantic distinction. One often-discussed case (see also Stuurman 1989;Menuzzi 1994;Kester 1996), involves non-intersective adjectives modifying nouns denoting societal roles, as in (8) Like veel A , één also has the -Ø ending in both the common and neuter singular indefinite (it does not occur in the plural), but -e appears in the definite (Booij 1992;Haeseryn et al. 1997). Menuzzi (1994) explains cases such as (8)/(9) by assuming that the adjective on the intended reading is merged higher than the functional head responsible for gender, but below Num and D, so that only the latter two can trigger agreement on the adjective. We could accommodate veel A by extending this analysis slightly so that veel A is generated higher than Num, but below D. Schoorlemmer (2009) does not discuss Dutch irregular adjectival inflection; the simplest extension seems to be that the vocabulary insertion rules for veel A spell out any specified value for gender or number as -Ø, and the elsewhere -e appears when the definite determiner blocks DP-internal agreement. Yet another option is to adopt the spell out rules of Menuzzi (1994) and Kester (1996), and postulate that veel A only has [udef], not [uNum] and [ugender]. I conclude that what I propose can readily be accommodated in existing theories of (irregular) adjectival inflection in Dutch; as it is not the purpose of this paper to decide on the choice between these theories, I will leave the matter open.
Turning now to vele Nl , nothing as yet forces us to abandon its customary analysis as an adjective, but we will encounter several indications later on that it is best characterized as a vague numeral. In order to describe the data seen so far, I retain Kester's assumption that vele Nl does not combine with mass terms; I return to the cause of this in section 3.3. This blocks the inflected vele Nl in (5a). (11) shows that the other vague numerals in this class also do not combine with mass terms: Even een enkel(e) in (11c)-(11d), which can appear with a grammatical singular (showing the -Ø ending with indefinite neuters in (11d)), nonetheless coerces mass terms to a nonmass reading (cf. English many a wine). 7 We find the same, familiar, pattern with cardinal numerals, which can combine with mass nouns in the plural, and sometimes even in the singular, but always coerce a non-mass reading: Since the pure measure reading is allowed here, what appears in (13a) must be vele Nl , so we can conclude that vele Nl can be preceded by a determiner, hence is not itself a determiner. Instead, as a vague numeral it can be preceded by a determiner, a property it has in common with the other (vague) numerals in (13b) and (13c). (13d) supports the same conclusion. Table 2 summarizes the inflectional properties I attribute to veel A and vele Nl .
See section 2.4 below for a summary of the evidence provided both here and in subsequent sections in favor of analysing veel A as an adjective and vele Nl as a numeral.

Veel A as a gradable adjective
The preceding section yields one argument that veel A behaves as an adjective rather than a determiner: it can be preceded by a determiner (see (6)). 8 This section presents evidence that the adjective is gradable.
To begin with, my reanalysis of veel A and vele Nl partly solves a problem noted earlier in the literature. As Broekhuis (2013) (14) is a (relative) gradable adjective (see Kennedy & McNally 2005 for discussion of the licensing of degree 8 I pass over the predicative use of veel, which also seems to support my reanalysis: (i) a. Dat is veel / *vele. that is much 'that's a lot' b. Dat is weinig / *weinige. that is little Recall that on the traditional analysis undeclined veel A is a determiner which is not expected to occur in this position. But declined adjectival vele Nl is supposedly distributive so it should not predicate over a (mass) subject in the singular. On my analysis, veel in (ia) can be the undeclined adjectival non-distributive veel A that also occurs in (5a) and (6). I cannot address the restrictions on predicative veel A here, or why it appears to force the subject to be mass.   (15) is not a (gradable) adjective but a vague numeral that does not take a degree argument, so that like the other numerals in its class it does not allow a degree modifier: (16) *nogal meerdere / ettelijke / luttele / enkele / verschillende boeken rather several / many / few / some / various books A somewhat problematic consequence is that we predict that a degree modifier should be allowed in combination with a declined form vele when it is preceded by the definite article, as this could be the declined form of adjectival veel A . Broekhuis (2013) presents data that contradict this (his judgment): (17) *de heel vele problemen the very many problems However, as an anonymous reviewer observes, there are unexplained and idiosyncratic differences here among degree modifiers. While I agree with Broekhuis' judgment in (17), I find (18a) less marked and (18b) even better (both are judged ungrammatical by Broekhuis; I agree that these examples are marked, but not much more than those in (6) and (7), where the mass noun forces a declined form of veel A ). Furthermore, (18c), provided by the anonymous reviewer, is completely well-formed.  (14) -for the gradability of veel A . Unfortunately, two other common tests for scalarity cannot readily be used to distinguish veel A and vele Nl : for some reason, neither variant allows a measure phrase, and one cannot easily tell from which variant the suppletive comparative and superlative forms derive. However, recent research on other types of modification made available by the semantics of degree expressions provides us with additional evidence that veel A is gradable, and vele Nl is not.
First, relative gradable adjectives allow modification that helps to specify the Comparison Class (see Bylinina 2013 for a recent overview of the literature). In (20a), for an 8-year-old indicates that Vera reads books that are lengthy compared to the books 8-year-olds generally read (see Solt  Likewise, (20b) indicates that the number of books Vera reads exceeds the expected number for 8-year-olds. This is exactly as expected if veel A is gradable. It also provides a modicum of evidence that veel A is an adjective, not a quantifier or a determiner, which are not ususally treated as gradable (an exception is Hackl 2000, who treats English many as a gradable GQ determiner, type <d,<et,ett> >, but I am not aware of evidence that it functions syntactically as a determiner). (20c) and (20d) show that vele Nl patterns with determiners, quantifiers, cardinals and vague numerals in not allowing this type of modification; I take (20c) to entail at least that vele Nl is not a gradable adjective (it does not appear plausible that it should be a gradable but absolute adjective). The traditional analysis of veel A as a quantifier does not predict the well-formedness of (20b); because of the unfamiliar concept of a distributive adjective, it is impossible to tell what the traditional analysis would predict for vele Nl in (20c). The scalarity of veel A is also reflected in its judge-dependence (see Saebø 2009 andagain Bylinina 2013). Unlike vele Nl (and similar quantifiers and numerals) veel A may appear embedded under a 'subjective' attitude verb. Saebø observes that, next to predicates of personal taste and certain modal verbs, this context allows dimensional adjectives, a fact he attributes to the presence of a judge parameter introduced by the covert POS morpheme that accompanies such an adjective in the positive. Hence, we can explain the acceptability of (21a) and (21b) by analyzing veel A as a gradable adjective, accompanied by a POS morpheme, as proposed in section 3.3 below.  vele Nl I find that Lisa many/3/all/several books reads 'I feel that Lisa reads many/3/all/several books' Finally, the gradable adjective veel A also occurs with the Nominal AIC construction in (22) analyzed by Fleisher (2008), with its typical flavor of "inappropriateness": a. Dat is een dik boek om aan een eerstejaars-student voor that is a long book comp to a 1st-year student prt te schrijven. to assign 'that's a long book to assign to a 1st-year student' i.e., 'that book is so long that it is inappropriate to assign it to a 1st-year student' b. Dat zijn veel A boeken om aan een eerstejaars-student voor that are many books comp to a 1st-year student prt te schrijven. to assign 'those are so many books that it is inappropriate to assign them to a 1st year student' c. #Dat zijn *vele Nl /drie/meerdere boeken om aan een eerstejaarsthat are many/3/several books comp to a 1st-year student voor te schrijven. student prt to assign not: 'those are so many/3/several books that it is inappropriate to assign them to a 1st year student' Fleisher argues that the infinitival relative clause in (22a) contributes a modal component to the calculation of the standard relative to which a book would count as lengthy. Again, veel A in (22b) patterns with other gradable adjectives. (22c) shows that (vague) numerals do not support such an interpretation (vele Nl is excluded independently because it does not appear in a predicative position).
Returning briefly to the topic of the previous section, we can now employ the distribution of comparison class PP's to confirm its findings: Since vele Nl does not license such PP's, vele in (23) must indeed be an inflected form of veel A in a definite DP, which is what I assumed above in order to explain that the inflected form can combine with a mass noun in the definite.

Intermediate summary
We have established in section 2.3 that veel A is gradable. This would have been surprising given the traditional lexical categorization of veel A as a quantifier in the category of iedere 'every' or as a determiner, since such elements are usually not gradable. Hence, the presumed category of veel A might have cast doubt on our contention that it is gradable; this problem has been removed in section 2.2, where we have argued that mass-compatible gradable veel A can occur to the right of a determiner (see the data in (6) and (23)), hence is plausibly an adjective, for which gradability is unsurprising. We have provided no further evidence that veel A is an adjective rather than some other category with a similar distribution in the DP; however, categorizing it as, e.g., a vague numeral would again render its gradability exceptional, as well as its compatibility with mass terms, so I will tentatively maintain that veel A is an adjective. This is not crucial here: my explanation of the pseudo-partitive data that are central to this article will depend on the gradability of veel A , not on its syntactic category. An anonymous reviewer presents one argument that suggests that, at least in one respect, veel A behaves more like a numeral that an adjective. It is commonly assumed (see, e.g., Haeseryn et al. 1997) that noun-pronominalization with Dutch "quantitative" er 'there' can occur with numerals, but not with adjectives, as (24)  If so, the acceptability of veel A in (25a) would indicate that it is more like a numeral. However, the ban on adjectives in this construction is not absolute, witness cases like (25b). I conclude that the argument is inconclusive (while noting that, if correct, the argument would also contradict the traditional categorization of veel A as a quantifier or determiner), and I will leave the matter for future research.
Turning to the category of vele Nl , we have seen that it, too, can follow a determiner (see (13)), hence is not itself a determiner or quantifier. While I have no positive evidence that it is not a (non-gradable) adjective, it patterns consistently with other (vague) numerals: it can follow a determiner; it does not allow degree modifiers or other indicators of gradability (see section 2.3); it is incompatible with mass terms (see (5a)); and it combines with measure nouns (see section 3 below). I will also argue in section 3.1 that vele Nl acts as a probe for agreement inside the DP, behavior which it shares with other numerals but which sets it apart from adjectives. The syntactic categorization of vele Nl is not crucial for my purposes here, but we will see that attributing to vele Nl a semantics similar to that of other numerals will allow us to describe its compatibility with measure nouns in pseudo-partitives.
for the traditional analysis. I will then briefly review the standard assumptions on the structure of Dutch pseudo-partitives. On the basis of this structure I will present a proposal that not only correctly derives the semantics of the construction, but also explains the distribution of veel A and vele Nl . I will focus almost exclusively on pseudo-partitives with true measure nouns such as liter or kilo.
Dutch has two subclasses of measure nouns (Klooster 1972): those that show singular morphology when preceded by a numeral larger than one, and those that show normal number morphology. 11 The former class is instantiated by liter in (26) We observe that veel A and vele Nl differ from cardinal numerals in that they require plural marking on liter (see Doetjes 1997: 190ff). We also observe that veel A does not allow the pure amount reading, but vele Nl does. The following data confirm that the form vele that occurs in the pseudo-partitive construction with the pure amount reading is indeed the numeral vele Nl , not some exceptionally inflected instance of the gradable adjective veel A . The form vele in (28), which allows the pure amount reading, is incompatible with the gradability modifiers discussed in section 2.3, indicating that is it indeed vele Nl , not veel A .
Replacing vele Nl with veel A in (29) allows the gradability modifiers but blocks the pure amount reading. (Note that the #-signs in (28) reflect unacceptability because of the gradability modifiers; the # in (29)  In conclusion, vele Nl allows a pure amount reading with measure nouns, but veel A does not.
How can these data be explained? The traditional analysis of veel A and vele Nl does not help to explain the data in (27) through (30). The supposed distributivity of vele Nl does not predict that it allows the amount reading in (27d) and (30b). 12 And the only semantic property attributed to veel A , that it does not need to be distributive, gives no clue as to why it does not allow an amount reading in (27b) and (30c). Note also that the mass/ count distinction does not capture the pattern in (27)/(30): both veel A and vele Nl can operate in the count domain, and it is veel A , the variant that can combine with mass nouns, that is blocked in the (27b) and (30c), which could be argued to be mass contexts. If anything, one would expect the reverse pattern.
At first glance, my analysis does not fare much better. On the positive side, the assimilation of vele Nl to the vague numerals remains intact, as the other vague numerals also force the plural morphology and allow a pure measure reading: (31) a. *meerdere / verscheidene / luttele liter sg wijn b. meerdere / verscheidene / luttele liters pl wijn several / various / few liter(s) wine 'several/various/few liters of wine' amount However, the contrast between the mass terms in (5), (11) and (12), and the measure nouns on their pure amount reading in (26), (27), (30) and (31b)  We can already observe at this point that an explanation in terms of number marking will not work. We could describe the distribution of vele Nl (and other vague numerals, with the possible exception of een enkele 'a small number') by postulating that it combines only with grammatically plural NPs, but this description does not extend to the cardinals. Cardinals can combine both with grammatically singular nouns (notably with measure nouns in (26b) and also with some mass nouns in (12c)) and with nouns with plural marking, so the contrast between (26b) and (26c) cannot be due to the presence of the morphological plural as such. More importantly, restrictions on grammatical number marking cannot be used to describe the distribution of veel A in (27) and (30), as it can combine with both grammatical singulars (5a) and plurals (3a).
In the following three sections I will propose an explanation for these observations. I will start in section 3.1 by briefly reviewing the standard assumptions on the syntax of Dutch pseudo-partitive constructions. This section also presents the explanation I adopt for the number marking facts in (26) and (27). In section 3.2 I will propose a semantics for measure nouns that is compatible with this syntax, and which makes it possible to state semantic generalizations that govern the distribution of veel A and vele Nl in (27) -(30). In section 3.3, I will consider possible underlying motivations for these generalizations. Section 3.4 briefly considers how far a non-standard syntactic analysis of Dutch pseudopartitives could go toward explaining the relevant data.

Constituency in the Dutch pseudo-partitive
In the literature on English pseudo-partitive constructions it is often assumed that five meters in five meters of yarn is the same measure phrase that appears in five meters tall, forming a constituent to the exclusion of the substance noun (of) yarn. For instance, Schwarzschild (2006) places five meters in the specifier of a QP dominating the NP headed by yarn. The standard assumption on the structure of Dutch pseudo-partitive constructions, however, which I will adopt, is that the measure noun (meter) takes the substance noun (yarn) as its complement, the two forming a constituent to the exclusion of the numeral (see, for instance, Van Gestel 1986;Van Riemsdijk 1998;Vos 1999;see Hankamer & Mikkelsen 2008 for a similar analysis of Danish). I will refer to this as the head-complement analysis, and to the alternative that treats five meters as a specifier, as the specifier analysis. The head-complement analysis and the specifier analysis are schematically illustrated for vijf liter water 'five liters (of) water' in (32a) and (32b), respectively.
I will briefly review some of the standard arguments for the head-complement analysis of Dutch pseudo-partitives, present some additional arguments, and then outline the account I adopt for the number marking data in (26) and (27) above. Van Gestel (1986) provides syntactic evidence that Dutch cardinal numerals are nouns that take a nominal complement (as had been argued for English by Jackendoff 1977), and he shows that this analysis also extends to pseudo-partitives: the measure noun heads its own DP and takes the NP headed by the substance noun as a complement. One point of evidence is that gender on the DP is determined by the measure noun, not by the mass noun: 13 (33) a. die C éne / halve / twee liter C water Nt that one / half / 2 liter water 'that 1/half/2 liter(s) of water' b. het Nt onsje Nt cocaïne C the metric.ounce-dim cocaine 'the little ounce of cocaine' c. dat Nt jaar Nt vakantie C that year holiday 'that year of holiday' Also, gender on the complementizer of a relative clause is determined by the measure noun: (34) een liter C water Nt die C / *dat Nt we gedronken hebben a liter water that we drunk have 'a liter of water that we drank' Van Gestel explains this by assuming that liter selects too low an (extended) projection of N as its complement for it to allow adjunction of a relative clause, so the relative clause must be attached to the projection headed by liter. These data are difficult to capture if two liter, etc., is syntactically a specifier or modifier.
An additional argument for the standard head-complement analysis starts from the observation that measure phrases headed by (singular) measure nouns, like other (singular count) NPs, cannot appear bare but require an indefinite article or numeral: die tas weegt *(een/drie) kilo that bag weighs a/three kilo 'that bag weighs a kilo/3 kilos' But observe that the numeral may and the indefinite article must be absent when the measure phrase appears inside a pseudo-partitive, in case the pseudo-partitive as a whole has another determiner: (36) a. Jan's (*een) (drie) liter wijn J.'s a three liter wine 'John's liter/John's 3 liters of wine' b. deze (*een) (drie) liter wijn this a three liter wine 'this liter/these 3 liters of wine' Now suppose that we adopt the specifier analysis. We cannot assume that John's or this in (36) is part of the supposed measure phrase specifier (cf. *the bag weighs John's kilo/this kilo, *John's meters tall). And this is indeed blocked if the measure noun that heads the measure phrase specifier is taken to require (as in the semantics of Krifka 1990, discussed below) a cardinal as an obligatory argument (with the indefinite article in (35) perhaps an optional variant of one), or on the semantics proposed in Schwarzschild (2006). However, this means that the supposed measure phrase specifier in (36) consist just of the noun liter, only optionally preceded by a numeral. It is unclear on the specifier analysis why the numeral may be missing, and why the indefinite article cannot appear inside the measure phrase specifier in (36), unlike in (35). On the structure adopted here, these data are unproblematic: the measure noun liter in (36) heads its own DP, which allows the same range of determiners as other similar DPs: a possessive or demonstrative in complementary distribution with the article, optionally followed by a numeral. The semantics proposed in section 3.2 will deal correctly with the definite determiners in (36).
The following observations also render the standard head-complement analysis more plausible than the specifier analysis: (37) a. dat Nt éne / *één jaar Nt oponthoud Nt that one year delay 'that one year of delay' b. die C éne / ??één liter C wijn C that one liter wine 'that one liter of wine' c. dat *éne / één jaar Nt lange oponthoud Nt that one year long delay As shown in (10) above, the numeral één 'one' is inflected when it appears in a definite DP. (37a) and (37b) show that this also obtains when the numeral precedes a measure noun. This is unexpected if éne is embedded in a separate measure phrase specifier éne jaar / éne liter. Observe, for instance, that the numeral één 'one' when it appears in a measure phrase modifying an attributive adjective does not agree for definiteness with the DP (see (37c)). These observations follow immediately on the analysis adopted here. In addition, adopting the head-complement analysis will allow a unification of the pseudo-partitive construction under discussion here with the group noun and container noun constructions exemplified in (38) (see Vos 1999 for an overview): (38) a. een groep toeristen a group tourists 'a group of tourists' b. een doos koekjes a box cookies 'a box of cookies' It is semantically implausible that a group or a box in these constructions should function as a measure phrase, at least on the reading where they entail the existence of an actual group or box. It follows that such nouns must be capable of taking a nominal complement, and providing it with case; this makes it more plausible that this also happens in pseudopartitives. Note furthermore that in previous stages of the language, the substance noun following a container or measure noun was visibly marked genitive (Stoett 1923: 102), as expected on the head-complement analysis (whereas in the five meters tall case, the measure phrase specifier would be marked genitive).
Thus far, I have argued that the measure noun liter and the substance noun water in pseudo-partitive vijf liter water 'five liters water' stand in a head-complement relation. In principle, this leaves open several options for the position of the numeral. Van Gestel (1986) argues that the same head-complement relation obtains here, as illustrated in (32a) above, an analysis I will adopt for the following two reasons. Firstly, because Ionin and Matushansky (2006) argue successfully that simplex numerals are nominal heads taking nominal complements; in a complex numeral DP such as two hundred books, books is the complement of hundred, and hundred books is the complement of two. They propose a corresponding semantics for cardinal numerals which facilitates a successful compositional treatment of complex cardinals. Secondly, because Matushansky and Ruys (2014) show that adopting this structure allows one to explain the puzzling pattern of number marking observed with measure nouns along the following lines.
Recall that some measure nouns remain in the singular when combined with a cardinal, and other measure nouns are pluralized (Klooster 1972); this is illustrated again in (39): (39) a. drie jaar sg vakantie three year vacation '3 years of holidays' b. drie maanden pl vakantie three months vacation '3 months of holidays' As I will argue in the next section, it is implausible that the plural marking on maanden 'months' in (39b) should reflect semantic pluralization, which does not occur in the mass domain. In addition, there is no relevant semantic distinction between year and month that could explain the contrast between (39a) and (39b) (see Klooster 1972 for some discussion). 14 Matushansky and Ruys (2014) conclude that number marking in these cases is a purely syntactic agreement phenomenon: pluralizing measure nouns like maand bear a syntactic feature [Ind] (for "individuation") that causes them to Agree with a probing cardinal numeral, triggering plural marking, while other measure nouns like jaar and liter lack this feature. 15 This is a plausible analysis only in case the cardinal numeral is a head that can probe into its complement and Agree with the measure noun.
Given this treatment of number marking, we can now assume that the vague numerals, including vele Nl , are also complement-taking heads, but are different in that they probe for some feature that all measure nouns share (say, N), causing all of them to pluralize. This will describe the number data in (26), (27), (30), (31) and (39): the vague numerals in (27d), (30b) and (31b) enter into an Agree relation with the measure nouns and cause them to be marked plural; the cardinal Agrees with the measure nouns in (30a) and (39b), but not with the measure nouns in (26b) and (39a). 16 See Matushansky and Ruys (2014) for further discussion. 17 This analysis warrants an additional conclusion. If vele Nl in (27d) indeed probes the measure noun and fixes its number feature, it becomes unlikely that vele Nl is a regular adjective. I am not aware of other adjectives that value φ-features on the nouns they modify; there are surely no adjectives that can be inserted before liter in (26a) or (26b) that will render the measure noun plural. This confirms my assessment in section 2.4 above that the limited evidence available suggests that vele Nl is not a determiner, nor an adjective, but indeed a (vague) numeral.
The analysis also entails that we cannot use grammatical number marking in (27) to detect the semantic number of the measure nouns. Whether or not we can use semantic number as the distinguishing property that allows vele Nl but not veel A in pseudo-partitives will therefore have to be decided by other, semantic considerations, which the next section will provide. I feel that one can conclude with a fair amount of confidence that the head-complement analysis for measure nouns and substance nouns in Dutch pseudo-partitives is correct, and with some confidence that the same holds for numerals and the NPs they combine with. I will argue in the next section that these assumptions can form the basis for an analysis of the semantics of these constructions that supports an explanation for the contrast between veel A and vele Nl in (27).

A semantics for Dutch pseudo-partitives
The syntactic analysis we have adopted for pseudo-partitives places restrictions on the kind of semantics we can adopt. It seems to me that the proposal in Schwarzschild and Wilkinson (2002) and Schwarzschild (2006), according to which the measure phrase three liters in three liters (of) wine denotes a predicate over intervals, cannot be employed if liters wine forms a constituent to the exclusion of three. Another option is the analysis in Krifka (1990), who builds up 200 liters of wine as shown in (40)  This analysis also combines liter first with 200, and then with wine, but this can easily be repaired by inverting the order of the arguments of liter. I cannot exclude that an analysis like this can be made to explain the contrast between veel A and vele Nl ; see section 3.4 below. Nonetheless, there are reasons not to adopt this treatment. One reason is that it is not compatible with the independently motivated semantics of cardinals from Ionin and Matushansky (2006). In addition, I have adopted the analysis that makes the numeral a head that takes the (measure) noun as a complement. Now if one takes the numeral as an obligatory argument of liter, it would be hard to understand why the numeral, and not the measure noun, projects when the two combine. More importantly, the treatment in (40) entails the obligatory presence of a (cardinal) numeral. I have argued on the basis 16 As to (27b), for reasons explained in section 3.2 below veel A can only combine with a predicate that has undergone semantic pluralization, presumably via an intervening Num head, which gives rise to the number marking and the non-measure reading. 17 Note that the plural in (27d) is not the plural of abundance referred to as the "emphatic" use in Hoeksema (2006); there is no abundance effect in (27d), or in its variant luttele liters wijn 'very few liters of wine'. of (36) that this is problematic: pseudo-partitives can appear with just a determiner, and no cardinal. 18 We can allow pseudo-partitives to appear without a cardinal and to combine directly with a determiner through a slight modification of the denotation of liter (cf. Lasersohn 2011: 1144): On this analysis, liter only takes the substance noun wine as a complement, yielding a (mass) predicate that applies to portions of wine of one liter. This predicate can combine with a determiner in the usual way. This explains (36). The next question is how we combine liters of wine with a cardinal to obtain three liters of wine, if three is not an argument of liter.
The most common treatment of cardinal numerals is as cardinality predicates: a cardinal combines with a semantically pluralized predicate and selects only those plural individuals that have the correct cardinality. However, this will not work without modification in the present case: if Link's (1983) standard operation of semantic pluralization were to apply to liters of wine, this would yield the set of all individual sums of one-liter portions of wine (not necessarily measuring multiple liters, since the original portions may overlap materially). Three, as usually defined, would select from these all three-membered i-sums, not just the three-liter sized ones. 19 To fix this, one would either need to define semantic pluralization in such a way that it only constructs i-sums of non-overlapping individuals, or define the cardinal so that it only selects plural individuals whose members do not materially overlap. Both solutions, though inelegant, as they are mandated only by the need to allow pluralization in the mass domain, are possible, but have the disadvantage that they no longer distinguish (42a) from (42b), since in (42a) we would also be dealing with an individual sum of three 1-liter units (see also footnote 19): (42) a. drie liter water three liter water 'three liters of water' amount b. drie liters water three liters water 'three 1-liter units of water ' unit We can avoid the complications that would arise from semantic pluralization of liters of water, and the problem raised by (42), by adopting the semantics for cardinals proposed by Ionin and Matushansky (2006). In this framework the non-overlap condition is inde-18 There is one additional plausibility argument against the analysis in (40): if liter requires a number as an argument, combining it with vele as in (27d) is problematic since vele does not denote a number (likewise for the other vague numerals; see (31b)). This problem could be solved by applying QR to vele, along the lines of (59) below; but I am not aware of evidence in favor of this operation applying here. 19 Such a reading is (marginally) available when we force the non-measure reading of liter by making it grammatically plural in combination with a numeral. Suppose we have a container containing 1.5 liters of water, from which we can drain the bottom 1 liter, or siphon off the top 1 liter. I cannot extract twee liter water from the container, but there are (marginally) twee liters water that I can extract.
pendently required, and cardinals combine with a semantic singular, as appears to be the case here. 20 We can then construct two hundred liters of wine as follows (starting from (41)): The partitioning operator can be defined as in (44)  To say that an individual x partitions into a set Y is to say that Y consists of non-overlapping individuals that together make up x. By (43), the predicate denoted by two hundred liters of wine is true of those individuals that can be partitioned into a set of two individuals, each of which can be partitioned into a set of 100 individuals, each of which is one liter of wine. If John buys one such individual, he buys two hundred liters of wine. Numerals combine with count nouns in the same way, in Ionin and Matushansky's (2006) approach, except that partitioning then functions in the count domain. In two books, two as defined in (43) combines with semantically singular books, which denotes the set of book atoms, and the combination yields a predicate that applies to plural individuals that consist of two books, i.e., that can be partitioned into a size-two set of non-overlapping individuals each of which is a book atom. 21 Now observe that predicates such as two hundred liters of wine or two books have the special property that the individuals they can be true of cannot stand in the proper part-of relation to each other. In this, they differ from predicates such as wine or (semantically plural) books, which can be true of (plural) individuals some of which properly contain others. Two hundred liters of wine or two books cannot apply to both x and y if x is a proper part of y, for the simple reason that these predicates apply to individuals that are all the same size (as measured by the measure function by which they are constructed): if x is two hundred liters of wine and is a proper part of y, then y must measure more than two hundred liters. Krifka (1990) calls such a predicate a quantized predicate (or a degree, as it can be used for measuring).
We are now in a position to describe the distribution of veel A and vele Nl and their kin. Consider again the data in (45) and (46): 20 Recall that Matushansky and Ruys (2014) argue that the number marking on measure nouns is an effect of agreement, so we can assume that the phrases liter wine in (26b) and liters wine in (27d) do not differ semantically: neither has undergone semantic pluralization. There is some independent evidence that liters wijn 'liters of wine' cannot undergo semantic pluralization: it cannot occur as a regular bare plural. The non-measure "liter-units" reading apart, liters wijn in (i) only has a reading as a plural of abundance: (i) Jan dronk liters wijn. Jan drank liters wine 'J. drank excessively many liters of wine' 21 Ionin and Matushansky assume, then, that plural marking on books in two books is also an agreement phenomenon, as proposed earlier by e.g. Krifka (1995), an assumption we have adopted here. Following Matushansky and Ruys (2014) We need to block the mass reading with vele Nl and other (vague) numerals in (46a), while allowing it with veel A in (45a). We need to allow both veel A and vele Nl with count nouns in (45b) and (46b). And we need to block the measure reading with veel A in (45c) and (45d), while allowing it with vele Nl and other (vague) numerals in (46c). I propose the generalizations in (47), for which I will be considering possible underlying causes in the next section: (47) a. Adjectival veel A cannot combine with a quantized predicate b. Vague numerals and cardinals can only combine with quantized predicates A mass noun does not denote a quantized predicate, so that the constraints in (47) allow (45a) but block the mass reading in (46a). 22 For the count nouns in (45b)/(46b) we assume, as is standard, that an NP that denotes a predicate over (count) atoms can optionally undergo semantic pluralization, presumably triggered by a functional head Num in its extended projection (Ritter 1991). As a result, boeken 'books' can either be semantically singular, so that it can combine with a (vague) numeral in (46b), since a predicate over (count) atoms is a quantized predicate, or it can be semantically plural (the result of semantic pluralization), so that it can combine with veel A in (45b), since a pluralized predicate is not quantized.
Recall that morphological number is not a reliable guide to semantic number here; in particular, we see plural marking on semantically singular boeken 'books' in (46b) because Agree with the probing numeral or cardinal triggers plural marking on the noun (see the discussion of Matushansky and Ruys 2014 in section 3.1). In (45b), morphological plural on boeken is presumably triggered by the semantically pluralizing Num head in the same way. Turning now to the pseudo-partitives, liter (on its pure measure reading) takes a mass noun to create a quantized predicate (as shown in (41)), which can therefore be input to a numeral or cardinal in (46c). This in turn creates another quantized predicate, allowing Ionin and Matushansky's composition of complex cardinals as in (43). Again, number marking in (46c) does not reflect semantic pluralization, and indeed occurs only with a subset of measure nouns and numerals (see above). But the same quantized predicate liter wijn cannot be input to veel A in (45c) or (45d), because of constraint (47a).
As for the non-measure readings observed: we can assume for (26c) drie liters wijn 'three one-liter units of wine' that it contains not a true measure noun but a container noun liter referring to actual liter units (e.g., bottles), which has the relevant feature [Ind] that makes it Agree with the cardinal, triggering plural marking (see Matushansky and Ruys 2014). For (45d) (=(27b)) it appears safe to assume that Num here has applied the semantic pluralization necessary to obtain a non-quantized predicate in conformity with (47a); since Num requires a set of (count) atoms this in turn coerces the same container noun reading (and Num triggers plural morphology); likewise for (30c). As for (45c), no interpretation is possible: we have seen that the pure measure reading is blocked by (47a); the non-measure, liter-unit reading is blocked because it would yield the required non-quantized predicate only after semantic pluralization via Num; but this would have caused morphological pluralization, as in (45d).
Let us briefly consider some additional cases with slightly different properties: We observe again that vele Nl combines with a quantized predicate, and veel A does not. The measure reading is blocked in (48a) because liters of marbles is quantized. It could become unquantized only by undergoing semantic pluralization but this operation only applies to sets of atoms, coercing the liter-unit reading (the substance noun knikkers 'marbles' on the other hand presumably is the result of semantic pluralization applying to knikker 'marble'). Likewise, semantically singular honderden mensen 'hundreds of people' (lit. 'hundreds people') is quantized (and, I assume, cannot undergo semantic pluralization), hence can be input to numerals such as vele Nl (or to three), but not to veel A . 23 Note finally that an alternative explanation of the distribution of veel A and vele Nl in pseudo-partitives in terms of semantic number does not seem plausible. Postulating that vele Nl , the other vague numerals, and cardinal numerals require semantically plural complements would correctly prevent them from combining with mass nouns but also, incorrectly, from appearing in pseudo-partitives, unless we modify the pluralization operation to add a non-overlap requirement, as discussed above. Also, it is at odds with Ionin and Matushansky's (2006)  would need to give up their compositional analysis of complex cardinals. Describing the distribution of veel A by restricting it to semantic singulars is even harder. To block (45d) one must then postulate that liter wine is obligatorily plural; it is not clear how this could be derived, and in view of (46c) it is again incompatible with Ionin and Matushansky's (2006) treatment of cardinals. 24

Motivating the semantic constraints
In this section I will propose a semantics for veel A and vele Nl , from which I will attempt to derive underlying motivations for the constraints in (47). The basic idea is that counting (with a cardinal or vague numeral) only makes sense for objects of the same quantity, and assessing relative quantity (with a gradable adjective) only among objects of different quantities. I have argued that veel A is gradable, presumably a gradable adjective. This makes available the following motivation for (47a). I largely follow Krasikova & Champollion's (2011) treatment of Russian mnogie 'many' as a gradable adjective (see also Hackl 2009), but I will gloss over many details irrelevant to the motivation of (47a). Let veel A denote a function from individuals to degrees, which assigns to every individual its degree of 'manyness', or its amount. Compare this to the denotation of tall (see Kennedy 1999 for this, and discussion of related treatments of gradable adjectives): In case x is a plural individual, I assume that amount simply returns the number of atoms in x. If x is a mass, the dimension measured depends on the kind and may also be judgedependent (but the measure function must be monotonic, see Schwarzschild 2002). A discussion of the source of the 'scale function' exceeds the scope of this paper; for our examples we can assume that the amount of a portion of wine is determined by its volume (in Solt 2015, the relevant function is provided by a functional head Meas whose value is context-dependent; see also Schwarzschild 2006 for discussion).
In the positive, the adjective tall or veel A combines with an abstract POS morpheme, which places the degree of height/amount yielded by the adjective above the standard height/amount. Since I am only dealing with attributive veel A , POS in (51) also takes care of combining the result with the denotation of the noun:

λx:N(x).A(x))(C)]
The standard of height/amount is calculated by a function std. Apart from the measure function for which the standard is calculated, this function also takes into account a contextually determined comparison class C; we can think of the comparison class PP's discussed in section 2.3 as (partly) determining C. Finally, the function takes into account the noun set that the adjective modifies (a tall man exceeds a different standard than a tall tower); this is built directly into attributive POS by restricting the domain of the measure function (as in Krasikova & Champollion 2011).
We obtain the following derivation for veel A mannen 'many men': This yields a predicate over those plural individuals of men whose cardinality exceeds the standard for the cardinality of plural individuals of men, taking into account the context (for instance, for a Tuesday afternoon).
The intuition I want to pursue as a motivation for why veel A does not combine with quantized predicates is that it makes no sense to predicate of an individual that it is relatively big among individuals that are all equally big. Consider the std function as it applies to veel A and mannen (or tall and man): it takes all plural individuals that consist of men and orders them by cardinality (or takes all men and orders them by height). For this range it then calculates the standard cardinality (height) by means of some statistical concepts (median and median absolute deviation, according to Solt 2011). Now consider what would happen in (45c), where veel A combines with the quantized predicate liter wine:

amount(x))(C)]
The contribution of POS veel A here is trivial by necessity. Std ranks all portions of one liter of wine by volume, and calculates a standard volume among these (carefully, if vacuously, taking the context into account). We then obtain the set of liters of wine whose volume exceeds this standard. Whatever the details of this procedure, the result is trivial: either we always end up with the same set of 1-liter portions of wine that we started with (if one liter exceeds the standard), or we always obtain the empty predicate (if one liter does not exceed the standard). Assuming that the standard is the median, the latter case obtains. 25 I submit that the triviality of modifying quantized predicates by veel A explains why it is unacceptable. The discussion so far has focused on the unmodified adjective veel A , so the semantics I have provided involves a role for the covert POS morpheme. This is not to say that POS is crucial in explaining the ban on veel A co-occurring with measure phrases; but with other instances of Deg, the calculation will be different. Consider,e.g.,te 'too' in (54): 26 (54) *Jan dronk te veel liter wijn. veel A Jan drank too many liter wine 25 The contribution of the context variable C cannot render the adjective non-trivial, in that it can restrict std to consider only a subset of the noun set (many men for a Tuesday afternoon calculates the standard on the basis of pluralities of men that appear on Tuesday afternoons), but cannot make std ignore the contribution of the noun set. 26 One might argue that (54)  The pure amount reading is excluded with te veel A 'too many' as well. We can understand why by considering that, based on our semantics for veel A , the sentence would assert, roughly, that the most voluminous liter of wine that John drank was more voluminous than the liters of wine he drank in any permissible world (see, e.g., Meier 2003;von Stechow et al. 2004 for details), which is vacuously false. 27 A similar reasoning can explain why the (suppletive) comparative of veel A is blocked. Likewise for #Jan dronk (een) erg veel(e) liter wijn 'John drank (a) very many liter wine', which would assert that John drank a liter of wine that was very voluminous. 28 A full derivation of such cases must await another occasion; what unites them and the POS case we have discussed in detail, it would appear, is that they involve some sort of predication (relatively large, too large, etc.) over the volume of a liter, which will generally turn out to be vacuous. 29 I will have less of interest to say about (47b). We have a little more leeway in how we deal with vele Nl ; since we have seen that it patterns with numerals, the obvious treatment is along the lines of the treatment of cardinals that I adopt from Ionin and Matushansky (2006) 27 Depending on technical details, perhaps it is true in one marginal case, namely if there was no permissible world where John drank a liter of wine. But saying that the liter he did drink was too large would be a pragmatically odd way of stating this, perhaps because it would implicate that a smaller liter would have been ok. 28 A special case is hoeveel 'how.many'. It appears to be composed of hoe 'how' and the uninflected veel A (as we expect, since the uninflected variant is gradable). If so, it would constitute a counterexample to generalization (47a), since unlike veel A it does combine with measure nouns to create a pure measure reading in hoeveel liter wijn 'how many liters of wine'. However, it rather appears that hoeveel is not composed of either veel A or vele Nl , since it combines with all three categories under discussion: mass nouns, count nouns, and measure nouns. That hoeveel in hoeveel liter wijn is not decomposable is confirmed by the fact that hoe 'how' actually cannot combine with regular adjectives in an attributive position, only in a predicative position: *hoe grote meisjes 'how big girls' vs. √hoe groot zijn de meisjes 'how big are the girls'. I will leave the proper analysis of hoeveel for future research. 29 In each of these cases, pluralizing liter will render the example well-formed, but only on the liter-unit reading, and the explanation is as before. Semantic pluralization with Num (triggering morphological plural) is possible for a container-noun 'liter', and will render liters wijn not quantized: we can then meaningfully predicate of these plural individuals of multiple liter-bottles that they are too large, relatively large, etc.
Obviously, to predicate over a portion of wine that it consists of many (more than one, three) portions of wine without stating the size of these portions does not provide more information than predicating simply that it consists of wine, so that the numeral is superfluous. See Chierchia (1998a;b); Ionin and Matushansky (2006) for discussion and references. The remaining cases of (47b) involve preventing numerals from combining with a semantic plural. Ionin and Matushansky (2006) do so by stipulating that (cardinal) numerals must select a set of atoms. If a numeral could combine with a plural, this would give rise to a somewhat absurd systematic and unresolvable ambiguity whereby three books, several books, and many books could have a reading of 'at least six books', 'at least four books', and 'at least twice the contextually determined number of books'; but I am not sure if this observation will serve to explain the restriction, so I will settle for stipulating (47b) or adopting Ionin and Matushansky's stipulation.

Measure phrase alternatives
My account of the contrast between veel A and vele Nl depends on the (standard) rightbranching syntax I assume for Dutch pseudo-partitive constructions. I have provided evidence for this analysis in section 3.1 above. Nevertheless, I briefly want to consider the question what kind of explanation could be devised if one assumes the analysis that takes the measure phrase five liters as a specifier in the extended protection of wine, as shown schematically in (32b).
Assume then that liter has the semantics in (40) from Krifka (1990): it combines first with a cardinal and then with the head (substance) noun. As pointed out by Schwarzschild (2002), this creates an immediate problem for the cases under discussion, such as many liters of wine, where the measure noun does not combine with a cardinal but with a vague numeral which does not denote a number. Unlike Schwarzschild, I believe there is a workable solution: following Solt (2015), allow many to undergo QR as shown in (59a), which could then be interpreted as in (59b)  It would be difficult to find independent evidence for QR taking place in these constructions. Indeed, one needs to appeal to the (unexplained) Heim/Kennedy generalization (Heim 2000) to prevent many from raising across other quantificational expressions, as the scope inversion this would result in is not attested. Also, movement of many in (59a) violates the Left Branch Constraint (cf. Kennedy & Merchant 2000). Nonetheless, for the sake of discussion let us adopt the QR solution, without which a specifier analysis along the lines of Krifka (1990) must be abandoned immediately. 30 With this solution in place, we can indeed employ the distinctions I have proposed between veel A and vele Nl to explain why only the latter can appear in pseudo-partitives. One way is to postulate that adjectives cannot undergo QR but vague numerals can, perhaps 30 An anonymous reviewer correctly points out that an analysis analogous to (59) will more easily allow one to derive a downward monotone reading for weinig(e), 'few, little'; clearly, an analysis on a par with the treatment of vele Nl in (55) does not lead to a monotone decreasing reading. Pending further research, however, it is my initial impression that the inflected form weinige, the apparent counterpart of vele Nl , tends to have a monotone decreasing reading by itself less easily than weinig, suggesting that (55) would be a better treatment than (59). Whether weinig(e) must indeed be treated exactly on a par with veel/vele, as the literature suggests, is a question I must leave for future research.
because they are more operator-like. Postulating a corresponding type distinction, which treats vele Nl as shown in (59b) above but veel A as a predicate (type <e,t> or <et,et>) will also prevent veel A from being interpretable in this construction. However, I am not sure how principled these explanations are. For instance, Solt (2015) treats both English much and many as gradable adjectives (predicates over sets of degrees) that undergo QR, which would support an analysis along the lines of (59); the problem of blocking veel A in pseudo-partitives while allowing it with count nouns then reappears. I conclude that the categorial and semantic distinctions I have claimed exist between veel A and vele Nl can support a technical solution for their distribution under the specifier analysis of measure phrases in pseudo-partitives, but the question whether such a solution can be given a principled basis must be left for further research. 31

Comparison with Russian and English
It is widely assumed, following Milsark (1974), that English many is ambiguous, its two instances differing both in their distribution and in their semantics. Russian has two overtly distinct instances of many (see Krasikova & Champollion 2011 and references cited there). In both languages, the syntactic difference appears to be that one variant is more adjectival, the other more determiner-like. Semantically the two variants give rise to a cardinal (weak) reading, and a proportional (strong) one. To conclude this article I want to provide a brief comparison of these Russian and English elements with veel A and vele Nl , in the hope that future research may successfully address the cross-linguistic variation observed, for which I have no account. There is a considerable body of work on the many readings of English many, and closely related work on Russian. Setting aside the issue of reverse proportional and related readings (Westerståhl 1985), English has been argued to have two instances of many. One behaves syntactically as an adjective (in that it can appear below a determiner) and is allowed in there-insertion contexts. It has a weak, intersective reading: (60a) states that the number of errors in your reasoning is high, not a high proportion of the total number of errors. Its semantics can be described as that of a cardinality predicate. The other many is disallowed in there-insertion contexts, but allowed as the subject of an individual-level predicate. It has a strong, proportional reading: (60b) states that the intelligent ones make up a large proportion of the theoretical physicists. It can be described as a strong GQ Determiner.
There are many errors in your reasoning. b.
Many theoretical physicists are intelligent.
Early discussions are in Milsark (1974) and Partee (1989); see Partee (2012) for a literature review and further references. More recent work discusses two instances of many in Russian: mnogie and mnogo (Babko-Malaya 1998). Mnogie is syntactically more like an adjective in that it shows adjectival agreement; mnogo does not. They also differ along the cardinal/proportional parameter but surprisingly, it is adjectival mnogie that has the proportional reading, whereas mnogo has a cardinal reading. Krasikova & Champollion (2011) describe the proportional reading for mnogie as resulting from a degree adjectival interpretation, where proportionality relative to the size of the noun set is mediated by the standard-setting function, as in (52)  Since I have proposed roughly the same semantics for veel A that Krasikova & Champollion (2011) propose for adjectival mnogie, this is what we expect: (61a) is proportional (the cardinality of the noun set 'physicists' is taken into account) because the cardinality of the noun set helps std set the standard for amount. Also, the semantics I tentatively proposed for vele Nl in (55) (3), and a cardinal reading seems perfectly acceptable for veel A in (3a) and in (62b). I do not have a firm proposal for dealing with this option; one possibility is that the standard for amount in (62b) takes into account not only the cardinalities of actual plural individuals of victims, but also cardinalities of victims in other possible worlds/contexts. But whatever the explanation, the data show that Dutch must be given a different treatment than Russian mnogie, which cannot appear in a context like (62) without triggering a marked reading. Also, Russian uses the non-adjectival mnogo for both measure nouns and mass nouns: (63) Russian a. mnogo viná much wine-gen b. #mnogie vína many-nom.pl wine-nom.pl 'many kinds of wine' c. mnogo litrov viná many liter-gen.pl wine-gen 'many liters of wine' d. #mnogie litry viná many liter.pl wine-gen.sg 'many 1-liter units of wine' Table 3 summarizes the distribution of many's in this three-language samplet (for historical reasons I use D to label non-adjectival instances):  Ruys: Two Dutch many's and the structure of pseudo-partitives Art. 7, page 30 of 33

Conclusion
I have proposed that Dutch uninflected veel is a relative gradable adjective (which inflects only for definiteness), and inflected vele is a vague numeral. This explains where (un) inflected forms appear in the DP and which forms allow degree modification, and leads to an effective semantic characterization. I have defended the right-branching analysis of Dutch pseudo-partitives and offered a compositional semantics for this construction that supports a natural account of which forms of veel/vele combine with mass nouns, measure nouns, and plurals. More work is required to obtain reliable data on the proportional/ cardinal distinction, and to address issues of cross-linguistic variation, which are receiving increasing attention.