Adverbial -s as last resort

This article examines the grammatical behavior of Dutch adverbs featuring so-called adverbial -s. This will be done on the basis of three questions: Firstly, what is the grammatical nature of adverbial -s? Secondly, in which structural configurations does it appear? Thirdly, what does adverbial -s tell us about the existence of adverbs as a separate part of speech? The article provides the following three answers to these questions: Firstly, adverbial -s is an affixal manifestation of the categorizing heads n and a (so-called -s-Support). Secondly, n and a externalize as -s when the raised root that forms an amalgam with the categorizing head is silent or a bound root. Thirdly, “adverbs” featuring adverbial -s are nominal, adjectival or adpositional expressions with an articulated syntactic structure. Some of these syntactic structures correspond to the so-called construct state. In short, linguistic expressions featuring adverbial -s do not support the idea that adverbs form a separate part of speech.


Introduction
I invite the reader to read the following short story written in Dutch: ( Traditionally, the italicized elements in (1) are characterized as adverbs. A conspicuous property that these Dutch adverbs have in common is their ending in -s. It is therefore not surprising that, in traditional grammar, this bound morphemic element is called 'adverbial -s' (Royen 1947(Royen -1954). Identifying -s as a marker signalling the part of speech 'adverb' suggests that traditional grammarians took these elements to have a composite structure. For example, the temporal adverb soms 'sometimes' in (1a) has the composite structure som + -s. The aim of this article is to further our understanding of this class of adverbs which, to my knowledge, has so far been largely ignored in the morphosyntactic study of Dutch (and other Germanic languages). 1 This will be done by addressing three questions: Firstly, what is the grammatical nature of adverbial -s? Secondly, what is the distribution of adverbial -s, that is, in which structural environments is it attested? Thirdly and most importantly from a broader theoretical perspective, what does adverbial -s tell us about the existence of adverbs as a separate syntactic category; that is, do adverbs constitute a separate syntactic category (i.e., a lexical category) within the parts of speech (i.e., besides N, V, A, P) or can they be reduced to other syntactic categories?
Before providing some initial answers to these questions, I think it is useful to point out that the question about the categorial status of adverbs has been on the generative-linguistic research agenda for a while. The tendency seems to be to place so-called adverbs in one of the other lexical classes. 2 For example, the adverb fast, as in John drove fast, can be related to adjectives, the temporal adverb yesterday, as in John cried yesterday, to nouns, and the adverb downstairs, as in John went downstairs, to adpositions (Bowers 1975;Emonds 1976Emonds , 1985Larson 1985;McCawley 1988;Jackendoff 1972). The reduction of adverbs to other syntactic categories is motivated by their (morpho-)syntactic behavior. For example, fast can be modified by degree words that typically combine with adjectives, as in very fast and how fast; the adverb yesterday can be followed by 's and act as a left branch constituent within a noun phrase, just like the nominal possessor John's. Compare, for example, yesterday's party and John's party. The fact that John's and yesterday's cannot co-occur within a single noun phrase suggests that they compete for the same position: *John's yesterday's party. Finally, the postnominal placement of downstairs in the restaurant right downstairs (is very comfortable), and its modifiability by right, corresponds to the distributional behavior and modifiability of PPs, as in the restaurant (right) down the road (is very comfortable).
The question about the categorial nature of adverbs includes the question about their syntactic compositionality; that is, to what extent do adverbs consist of smaller components? There is consensus that English -ly-adverbs such as carefully, as in John opened the door carefully, fall apart in two components: careful and -ly. It is generally assumed that the first component is an adjective. There is less agreement, though, on the grammatical nature of the bound morpheme -ly. It has been analyzed, among others, as an inflectional suffix on the adjective (Emonds 1985:201), as the lexicalization of a Relator-head that mediates in a subject-predicate-that is, VP subject AP predicate -relationship (Den Dikken 2006:30-31), and as a bound-morphemic noun that is modified by an attributive adjective (Déchaine and Tremblay 1996;Baker 2003). 3 In recent years, the perspective on lexical categories has changed radically. An approach to categories has been developed, which starts from the assumption that lexical categories (also called content words) as such do not exist as primitive elements. Rather they are roots, unspecified as to category. Their categorial status as nominal, verbal et cetera results from merger with a categorizing head (Marantz 1997;Embick and Marantz 2008). 4 For example, the root kiss becomes a noun by merging with the nominalizer n, as in n+ √ kiss, and it becomes a verb by merging with the verbalizer v, as in v+ √ kiss. 5 An important consequence of this analysis is that lexical categories are syntactic objects with a composite structure.
Having provided some background for the research carried out in this article, let me return to the three earlier questions related to the grammar of Dutch adverbial -s. I will provide the following answers to these questions: Firstly, adverbial -s is an affixal manifestation of the categorizing heads n and a. Its appearance (-s-Support) will be analyzed as a last resort phenomenon. Secondly, the appearance of adverbial 3 For further discussion of the composite structure of adverbs, see Alexiadou (2013), Delfitto and Fiorin (2017). 4 See also Borer (2005) for the idea that the categories 'noun,' 'verb' et cetera are configurationally defined syntactic objects. She does not adopt categorizing heads such as n and v but proposes that roots become nouns or verbs by merging with functional categories such as D or T, respectively. 5 In the linguistics literature, roots are generally represented in upper-case letters (e.g., n+ √ KISS). In this article, I will simply represent them by means of lower-case letters (e.g., n+ √ kiss) in order to distinguish them from silent (i.e., unpronounced) roots, which, in the spirit of Kayne (2003), will be represented by upper-case letters.
-s depends on the nature of the amalgam that results from raising of the root to n or a. Thirdly, adverbs ending in -s are syntactic objects with an internal articulated structure. Specifically, they are nominal, adjectival or adpositional syntactic objects. Thus, there is no reason for treating these -s-bearing expressions as belonging to a separate class of speech, namely, the class of adverbs. The paper is organized as follows: Sect. 2 introduces the phenomenon of adverbial -s by way of describing some of its distributional and morphosyntactic properties. A central claim will be that expressions featuring adverbial -s have an inner syntactic structure. In Sect. 3, it is proposed that adverbial -s is a last resort manifestation of the categorial heads n and a. In Sect. 4, these two major claims -the inner structure of -s-bearing adverbs and the last resort nature of -s-will be substantiated by means of three case studies. Each case study introduces a structural configuration in which -s appears as a last resort phenomenon: Sect. 4.1 examines the pattern zachtje-s (slow-DIM(INUTIVE)-s, 'slowly'), Sect. 4.2 the pattern namens (name-en-s, 'on behalf of'), and Sect. 4.3 the pattern ondergronds (under-ground-s, 'underground, subterranean'). Section 5 broadens the empirical scope by providing case studies on temporal adverbs (5.1), measure adverbs (5.2), and degree adverbs (5.3). Section 6 extends the discussion of adverbial -s to the adjectival domain. Section 7 is the conclusion. It briefly addresses the question as to what the present study on Dutch adverbial -s implies for the broader (i.e., cross-linguistic) study of so-called adverbs.
Having introduced some of the distributional properties of adverbial -s, I will now turn to some of its morphosyntactic properties. In line with intuitions of traditional grammarians, I take expressions featuring adverbial -s to have a composite structure. One might object to this and argue that these adverbs no longer have any internal structure in present-day Dutch, even though, historically, they arguably had. Under such an approach, these adverbs ending in -s would be non-decomposable lexical units. In what follows, I will give some initial support for the idea that these expressions have an internal syntactic structure, though. In Sects. 4, 5 and 6, a more detailed discussion of the internal syntax of these (and other) expressions will follow.
Let me start out with the observation that adverbs ending in -s display signs of structure, in the sense that discrete, meaningful units can be identified within the adverb. The fact that these units can sometimes alternate with other units-in/op in (4), gronds/hands and onder/boven in (5), na(a)m 'name'/kracht 'force' in (6) A second sign of structure comes from example (7). The adverb straks 'soon' clearly falls apart in pieces (i.e., strak + -s) when the diminutive suffix is part of the structure: (7) a. straks soon-s 'soon' b. strak-je-s soon-DIM-s 'soon' 6 The -en following nam-(3b) used to be a dative case in older Dutch (Van der Sijs 2010).
7 In (4)- (7), and also in later examples, I often use Dutch orthographic conventions, which means that the adverbs are written as single units. Importantly, these units have an inner syntactic organization.
A third sign of internal structure comes from phonological stress. Many adverbs ending in -s display a stress pattern that corresponds to the one found in syntactic phrases. They display a phrasal stress pattern-that is, stress on the most deeply embedded (overt) constituent on the recursive side (Cinque 1993)-rather than a compound stress pattern-that is, stress on the first element of a complex word. This is exemplified in (8), where phonological stress is indicated by small capitals:  (8b), which suggests that ondergronds has a phrasal structure, as in (9). As I will argue later, gronds can be further decomposed into smaller units: grond+-s. In (10), daar refers to kamers, a subpart of binnenskamers. The fact that this coreference relation is possible suggests that the expression binnenskamers has a phrasal structure rather than the structure of a complex word. A nominal subpart of a complex word typically does not enter into a coreference relationship with an external antecedent. This is exemplified in (11) A final sign of internal structure comes from the grammatical behavior of adpositional phrases featuring a preposition ending with adverbial -s. Specifically, this behavior concerns the phenomenon of R-pronominalization. As noted in van Riemsdijk (1978), a pronominal complement of an adposition occurs to the left of that adposition and converts morphologically into a so-called R-pronoun (i.e., a pronoun featuring the sound /r/). According to his analysis, a pronoun starts out in the complement position of a preposition (i.e., P + PRON; e.g. op wat 'on what') and moves to the specifier position of P, which yields the postpositional word order (i.e., [ PP PRON i [ P' P t i ]]). It is in the specifier position that the pronoun gets converted into an R-pronoun; for example, wat + op (what on) gets converted into waar + op (where on 'on what?'). An example is given in (12) (13) shows that tijdens does not display the grammatical behavior of a regular adposition. Specifically, leftward placement of the pronoun and subsequent R-conversion are impossible. From this non-adpositional behavior I conclude that tijdens should not be analyzed as a non-decomposable adposition. Rather, it has a composite structure (tijden-s). It is this more complex internal structure that accounts for the impossibility of having leftward displacement and subsequent R-conversion of the pronoun; see Sect. 4.2.
In sum, many so-called adverbs ending in adverbial -s have an internal organization. They are complex syntactic objects. One of the elements of the composite structure is the adverbial marker -s. In Sect. 3, I examine more closely the grammatical nature of this small element.

Adverbial -s as a last resort manifestation of categorial n and a
If adverbs featuring adverbial -s have an internal syntax, what is the grammatical nature of -s? In traditional grammar (Royen 1947(Royen -1954, it is noted that adverbial -s was a genitival case originally. Since genitival case appeared on a large number of words having an adverbial function, it was considered to be an adverbial suffix; that is, a suffix that turns a word of category Y into an adverb. For example, the noun dood 'death' in (1c) converts into an adverb after attachment of -s. Of course, fulfilling a certain grammatical function (in casu adverbial modification) does not imply that the word or phrase carrying out that function belongs to a particular part of speech. For example, the fact that the English word fast can be used adverbially, as in John drove fast, does not lead to the conclusion that it is a category belonging to the part of speech type 'adverb' (Emonds 1976(Emonds , 1987a. Clearly, fast can fulfill other grammatical functions, such as predicative complement (This car is fast) and attributive modifier (a fast car). Likewise, linguistic expressions carrying adverbial -s can have different grammatical functions, as illustrated in (14) for the Dutch word anders (other-s, 'different(ly)'). Next to being a manner adverbial, as in (14a) If a word's categorial (i.e., part of speech) status does not follow from its grammatical function within a larger syntactic configuration (e.g., a clause), we are back to our original question: What is the grammatical nature of -s? My answer to this question builds on the idea that lexical categories such as nouns, adjectives and verbs have the composite structure f+Root, where f is a categorial head (e.g., n, a, v) and the root is unspecified as to category (Marantz 1997;Harley and Noyer 1999;Borer 2005). Specifically, I propose that adverbial -s instantiates the categorizing head n, and, as we will see in Sect. 6, the categorial head a. Of course, if -s is a manifestation of the categorizing head n, the question arises as to when it surfaces. It clearly does not pop up in all nominal environments. For example, -s is impossible in (15a) but obligatorily present in (15b): (15) a. in in één(*-s) one keer(*-s) time 'at once' b. ineen* (-s) in-one-s 'at once' The contrast in (15) suggests that the appearance of -s correlates with the absence of an overt root. When there is an overt root (keer), the appearance of -s is blocked, as in (15a). On the contrary, when there is no overt root, -s must surface, as in (15b). In Sects. 4-6, I will discuss various nominal and adjectival patterns in which adverbial -s surfaces. Before exploring each of them, I think it is helpful to introduce already here the abstract configurations that feature -s. But let's first briefly discuss the pattern in which -s does not surface, as in (15a). A basic assumption, already mentioned before, is that nouns have the base form n+root (= √ ), where 'root' starts out as the complement of n. The noun is "created" (derived) by head-moving the root to the categorizing head n, which I take to be an affixal element. The amalgam [root+n] is conventionally taken to be a complex head in which the raised root is adjoined to n, yielding the complex head [ n root+n]. To make things more concrete: head-movement of √ keer to n in (15a) yields the complex head [ n √ keer+n]. Following Chomsky (2015:12), however, I depart from this conventional analysis of the complex head and will assume that the affixal categorizing head n is adjoined to the raised root. 8 I will further adopt Chomsky's proposal that the root, even though being the host of the affix, cannot label, and that it is the amalgam 'root+categorizer' that plays a role in labeling (the projected structure). As argued in Chomsky (2015:12), the categorizing affix adjoined to the root is invisible to the labeling algorithm, which determines the properties of the syntactic object that results from merger of the root with the categorizing affix (Chomsky 2013). Building on this proposal, I propose that n, being an affix adjoined to a (freemorphemic) root (e.g., √ keer), is invisible for last resort Spell-out at the Syntax-PF interface. Consequently, the categorizing affix n does not externalize as -s, whence the ill-formedness of in één keer-s (see (15a)). 9 Let us next turn to the configurations in which the categorial head n does surface as -s. I propose that externalization of n as -s relates to the nature of the amalgam root+n, and especially to the nature of the root. Specifically, I take there to be two contexts in which the categorizing head must surface. Firstly, n externalizes as -s when the raised root is silent (e.g., √ TIME), as in (15b). 10 The intuitive idea here is that, in a complex head, at least one of the subparts must externalize at PF. 11 If the root does not materialize (i.e., spell out overtly), then n must: This means that the appearance of -s in a structural context like (15b) is a last resort operation (Chomsky 1986(Chomsky , 1995. Let's call it '-s-Support.' One might want to relate the need to insert -s to the principle of recoverability: -s must be inserted in order to make the information that we are dealing with a nominal construct (i.e., nP) formally recoverable at the sound surface. The second configuration in which n externalizes as -s is one in which the raised root is a Bound-Morphemic element instead of a free morphemic element. In other words, we have the following amalgam: [ √ BM + n affixal ]. 12 I assume that in this exocentric head-head configuration, where the subparts share the property of being bound morphemes, the two heads must conjointly externalize at PF. 13 In a way, their mutual dependence-both being bound-morphemes-requires the two elements to materialize together, yielding a single lexical unit at the surface. Also here, I take externalization of n as -s to be a last resort operation: -s-Support must take place in order to externalize the exocentric complex head. 9 I assume that the ill-formedness of in één-s keer (see (15a)) is due to the fact that the root (keer) has not raised to affixal n. If the root does not raise, there is no host for n to attach to. Furthermore, the root, remaining in situ and consequently being uncategorized, is an uninterpretable element in the syntactic representation. As noted by Chomsky (2015), raising of the root to a categorizing head is a universal property of language. 10 Importantly, I assume that the amalgam [ √ TIME+n] is a legitimate (i.e., interpretable) construct at the syntax-LF interface. Thus, realization of n as -s is a phenomenon at the syntax-PF interface. It should be noted that -s does not seem to contribute any meaning to the adverbial pattern in which it is embedded. 11 As an alternative, one might consider an analysis in which -s spells out the entire complex head, that is: As will become clear soon, there are patterns in which n spells out when the root has phonological contents. In those cases, it is clear that -s externalizes a subpart of the complex head. I will therefore assume that it is always the categorizer n that materializes. 12 The existence of bound roots is familiar from examples such as English √ struct, as in instruct and construction. 13 In a way this is reminiscent of Chomsky's (2013) proposal that in an exocentric XP-YP-configuration, X and Y can be taken as the label of this syntactic object if X and Y are identical in a relevant respect (e.g., by sharing a certain feature).
In order to make things more concrete, let me give an illustration of this second configuration in which -s-support takes place, namely the indefinite pronominal iets 'something' (see Sect. 5.2). This pronominal has a composite structure: iet + -s. The first part never occurs on its own in present-day Dutch: *iet. I take this element to be a bound root which, after being raised to (affixal) n, creates the following complex exocentric head: [ √ iet BM + n affixal (= -s)]. Summarizing, I have proposed that the appearance of adverbial -s depends on the nature of the root that merges with n. A raised free root (e.g., √ keer) yields a complex head in which the root constitutes the host for categorial n. A raised silent root, as in ineens (15b), yields a complex head in which n externalizes as -s as a last resort. Finally, a raised bound root (e.g., iet) yields an exocentric complex head, whose subparts, being bound morphemes, must externalize conjointly, where externalization of affixal n is a last resort operation.
So far, I have argued that -s is a last resort realization of the categorizing head n (and also a; see Sect. 6). A question which has not been answered yet is the following: Why -s as an externalization of n (and a)? My answer to this question starts from the traditional grammarians' claim that so-called adverbial -s relates to genitival case. Specifically, I build on Emonds's (1985Emonds's ( , 1987b proposal that genitival caseand Case, more generally-is not an independent, primitive category but rather an affixal realization of a part of speech (see also . In Emonds's formulation, Case is an alternative realization of the categorial head that selects a noun phrase as its complement. For example, a noun marked for genitival case, as in Latin domus puellae (house girl-GEN, 'the girl's house') is a noun carrying a nominal suffix, here represented as 'N': [ N domus] puell[ N -ae]. 14 Although genitival case (i.e. Noun affixal ) normally surfaces ex situ on a satellite constituent of the noun, I propose that it can also surface in situ as a last resort strategy. Under the above-mentioned assumption that nouns have the base form n+root, genitival case equals affixal n. Socalled adverbial -s is then a last resort in situ realization of affixal n (and affixal a, as we will see in Sect. 6). 15 Having shown that adverbs featuring adverbial -s are syntactic constructs with an inner structure (Sect. 2) and having proposed that adverbial -s is a last resort manifestation of the categorizing head n, I now turn to three case studies on adverbial -s. 14 The element -s no longer shows up productively on a satellite constituent of the noun. There is a small set of nominal expressions, though, in which a noun appears to assign genitive case to its complement, as in possessive noun phrases such as de heer des huizes (the lord the-GEN house-GEN, 'the master of the house') and de tand des tijds (the tooth the-GEN time-GEN, 'the test of time'). These are, however, fixed idiomatic expressions. An expression like *de fiets des jongens (the bike the-GEN boy-GEN, 'the boy's bike') is impossible in present-day Dutch. For a brief remark on the use of -s on prenominal possessors, as in Jans fiets (Jan-s bike, 'Jan's bike'), see Sect. 7. 15 See Rouveret and Vergnaud (1980), Chomsky (1981Chomsky ( , 1986 for the idea that nouns (N) and adjectives (A) are the syntactic categories involved in assignment of genitival case.

Adverbial -s in manner adverbial expressions
The first case study on adverbial -s concerns manner expressions such as zachtjes in (16a). As the gloss indicates, three components can be identified in this type of adverbial expression: the adjective (zacht), the diminutive morpheme -je, and adverbial -s. Note that the appearance of -s is obligatory. The above-mentioned data suggest that the appearance of adverbial -s and the inner morphosyntactic behavior of the adverbial expression are rule-governed. With these data as our empirical basis, let us next address the question as to what the internal syntax of these expressions is. As already noted, the diminutive morpheme -je is best known for its suffixal attachment to nouns, as in matras-je (17) and tafeltje (table-DIM, 'small table'); see Wiltschko (2005), De Belder (2011). Attachment of the diminutive morpheme to the noun does not change the categorial status of the newly built complex construct. 19 Thus, tafeltje constitutes a nominal construct, just 18 Certain speakers of Dutch permit the form netjeser (net+DIM+s+COMPAR, 'properly/decently'). They often also permit netjes as an attributive adjective: een netjes iemand 'a decent person'. Possibly, the diminutive morpheme in this pattern is a "low" diminutive in the sense of De Belder et al. (2014); that is, a derivational suffix involved in word syntax rather than a functional head involved in phrasal syntax. 19 Attachment of the diminutive morpheme has an effect on gender in Dutch: nouns carrying the diminutive morpheme are [+neuter]. Thus the [-neuter] noun tafel 'table' (e.g. de tafel) becomes [+neuter]: het like tafel 'table'. This is further confirmed by the fact that it can be followed by a plural morpheme (in casu: -s), just like tafel: tafels (table-PL), tafeltjes (table-DIM-PL). From this grammatical behavior of the diminutive morpheme, it can be concluded that it is a nominal type of bound morpheme. This brings us to the next question: How does the diminutive morpheme combine with the adjective?
In line with Wiltschko (2005) and De Belder (2011), I propose that the Dutch diminutive marker -je is a functional head in the extended nominal projection. It contributes the meaning component 'smallness.' Given its quantity-related meaning, I label the projection headed by -je as QP, which I take to be a layer on top of nP. 20 Thus, matrasje (mattress-DIM) in (17) has the base structure in (24a). Its surface form is derived by first raising the free root matras to n (24b), and subsequently raising the amalgam [matras+n] to Q (-je), as in (24c). I added the representation in (24d) in order to show what the extended nominal projection looks like when QP is embedded in a larger noun phrase containing an attributive AP. As indicated, I assume that the attributive AP zacht occupies the Spec-position of a functional layer representing attributive modification (Cinque 1994 Let us now consider the structure of zachtjes. I propose it has the base structure in (25a) and the derived structure in (25b): According to this analysis, zacht is an attributive AP that acts as a modifier within an extended nominal projection whose "noun" (i.e. root) is silent. 21 From bottom to top, this projection consists of the following components: (i) a silent root encoding 'manner,' here represented as WAY, (ii) a categorizing head n that selects the 'manner' tafeltje. This gender change can be identified on the basis of the definite article: [-neuter, +singular] nouns combine with de 'the', [+neuter, +singular] nouns with het 'the'. Plural nouns always take de as their definite article. 20 Wiltschko takes the German diminutive to be a classifier heading ClasP. De Belder analyzes Dutch -je as a Size-head heading SizeP. See Sect. 5.2 (on the Measure Phrase ietsjes) for an argument that -je occupies Q. 21 My analysis of zachtjes is similar in spirit to Déchaine and Tremblay's (1996) analysis of English -ly adverbs (e.g., quickly). They propose that quick is an attributive AP that modifies -ly, which they treat as a noun. See Kayne (2003) for the existence of silent nouns. root, (iii) the diminutive -je heading the QP-layer, and (iv) the attributive adjective phrase zacht in Spec,FP. 22 As shown by (25b), the pattern zachtjes is derived by raising the silent root to n, which yields the amalgam [ √ WAY+n]. I assume that, in this complex head configuration featuring a silent root, the categorizing head n externalizes as -s as a last resort operation; that is, at least one of the subparts of the complex head must materialize at the syntax-PF interface. If the root does not externalize, n must. I assume that the complex head [ √ WAY+n], which features the silent root, does not raise and adjoin to the diminutive morpheme -je in Q, possibly because the amalgam [ √ WAY+n (= -s)] is phonologically to weak to act as a syntactic host for a diminutive suffix. 23 Since -je and -s are not united by means of syntactic head movement, I assume they are united by a post-syntactic merger operation in the sense of Embick and Noyer (2001). Specifically, the affixal diminutive element je undergoes string-vacuous (i.e., non-inverting) Local Dislocation, and concatenates with what follows, yielding the linear sequence je+-s, which constitutes a syllable: /j@s/. The analysis in (25) provides a straightforward account of the phenomena discussed in (20)-(23). First, it is expected that the adjective, being an attributive modifier in [Spec,FP], can be accompanied by a degree word, as in (20)-(21). Second, the pattern in (23b) is excluded for the same reason that the ill-formed noun phrase *een bange man voor kritiek (an afraid man of criticism, 'a man afraid of criticism') is, namely: a PP-complement can never be extraposed out of an attributive AP and be placed in the right periphery of the containing noun phrase. Schematically:  22 One may wonder whether the nominal expression in (25b) is embedded within a larger PP headed by an empty preposition, as has been proposed, for example, by Bresnan and Grimshaw (1978), Emonds (1976Emonds ( , 1987b and McCawley (1988) for English NP-adverbs such as my way (I did it my way) and last week (John left last week). Although it, certainly, is worthwhile to systematically explore the presence of this prepositional layer on top of the noun phrase, I will leave this issue for future research, due to space limitations. Let me give one argument, though, which seems to go against a PP-analysis for the manner adverb zachtjes. The argument comes from extraposition, that is the postverbal placement of a phrase in an (underlyingly) verb-final clause. Normally, PPs and CPs, but not NPs and APs, can occur in extraposed position in Dutch. For example, the PP op een slome manier 'in a sleepy way' can occur either preverbally or postverbally, as indicated (by < >) in (ia). As shown in (ib), zachtjes can only occur in preverbal (i.e., non-extraposed) position. This suggests that zachtjes is not embedded in a PP. Notice, by the way, that the bare temporal noun phrase die dag 'that day' can occur postverbally, which suggests that this phrase is embedded in a PP: Third, the ill-formedness of (22b) follows from the fact that the bound comparative morpheme -er cannot attach to a nominal element, as depicted in (27): (27) is ill-formed for the same reason that *een zacht matras-er (a soft mattress-COMPAR) is, namely: nouns do not act as hosts for a bound comparative morpheme. Comparative -er can only be hosted by an adjective: een zacht-er matras (a soft-COMPAR mattress, 'a softer mattress').
As for the ill-formed pattern *zacht-er-tje-s in (22a), I tentatively propose that it follows from the interaction between the synthetic-comparative form zachter and the QP -tje-s. 24 Specifically, I take the former to be derived by movement of zacht to the bound comparative morpheme -er (i.e., Q [compar] ) yielding the adjectival structure [ QP zacht-er [ AP zacht]] (Corver 1997). 25 Observe that this pattern ends with a trace (i.e., an unpronounced copy). I propose that the bound-morphemic amalgam -tjes, which needs an overt host to its immediate left, cannot attach to zachter (via morphological merger) and form a phonological unit with it due to the presence of the trace (of displaced zacht). Schematically: This account of the ill-formed pattern zachtertjes is reminiscent of the ungrammaticality of wanna-contraction in sentences such as *Who do you wanna get the wine? (Compare: Who do you want to get the wine?). As has been argued in the literature, contraction of want and to to wanna is blocked as a result of an intervening wh-trace (Selkirk 1972): (29) Who do you want who to get the wine? (contraction blocked) The pattern minder bleekjes in (21) is well-formed for the simple reason that bleek remains in situ. The free comparative morpheme minder occupies (adjectival) Q, and, therefore, zacht does not raise and adjoin to Q. Since zacht remains in situ, there is a lexical host to which the bound morpheme -jes can attach via morphological merger.
Summarizing, I have argued that the adverbial -s at the end of the pattern zachtjes is an affixal realization of the categorizing head n. The last resort operation '-s-Support' is induced by the silence of the root in the amalgam [ √ WAY+n].

Adverbial -s on "prepositions"
Consider the italicized expressions in (31): 26 24 Note that positive adjectives ending in -er (e.g., dapper, 'brave') can be followed by the diminutive morpheme, as in dappertjes (brave-DIM-s, 'bravely'). From this, it can be concluded that the impossibility of zachtertjes is not a purely phonological thing. The grammatical nature of comparative -er matters. 25  Traditionally, the italicized words are analyzed as prepositions featuring adverbial -s. In one important respect, though, these words do not behave like prepositions at all. As touched on briefly in Sect. 2, the complement cannot be pronominalized by a so-called R-pronoun and be placed before the adposition, yielding a postpositional pattern (van Riemsdijk 1978). 27 This is shown in (32a,b), which are the Rpronominalized variants of (31a,b).
(32) a. *Jan Jan The question arises as to what causes the ill-formedness of the patterns in (32). As already hinted at by the glosses in (31) and (32), the "prepositions" ending in -ens contain a nominal element: naam 'name', tijd 'time'. The presence of this noun suggests that these "prepositional" elements are not simplex but have a composite structure. This composite structure becomes plausible when we consider the italicized patterns in (34a,b), which are the periphrastic counterparts of the synthetic forms namens and tijdens: Given the fact that the periphrastic pattern and the synthetic pattern are quite similar in meaning and lexical make-up, it does not seem implausible to assign them the same underlying structure. Notice at this point that the periphrastic pattern features the element van, which marks a possessive relationship. In (34a), for example, the noun phrase de regering is the possessor and the noun naam the possessum. In the spirit of Longobardi's (2001) analysis of (the history of) French adpositional constructions featuring chez 'at' (e.g., chez Pierre, 'at Jean's place'), I propose that expressions such as namens de regering also encode a possessive relationship. Specifically, they instantiate the so-called Construct State (CS) pattern (Ritter 1988;Siloni 1996;Borer 1999), which is familiar from Semitic languages such as Hebrew. As shown in (35), Hebrew can express a possessive relationship in two ways, namely by means of a so-called free genitive construction featuring a dummy case marker shel, as in (35a), or by means of a CS-construction like (35b), which involves no such case marker.
(35) a. ha-bayit the-house shel of ha-mora the-teacher 'the teacher's house' b. beyt house ha-mora the-teacher 'the teacher's house' Besides the presence versus absence of shel, there is another characteristic that distinguishes the free genitive construction from the CS-construction, namely the morphophonological form of the possessum-noun. In (35a), we have the free form bayit, in (35b) the CS-form beyt, which loses its main stress. This formal distinction is mirrored by the Dutch constructions in (31a) and (34a): the latter construction features the (free) form naam, the former construction the form namen (= naam+-en), which cannot occur on its own. I propose that namen is a complex word consisting of the root √ na(a)m and the bound morpheme -en. More specifically, I take -en to be a "low" bound morpheme, that is, a morpheme structurally located in between n(P) and the root. Schematically: 28 I assume that the categorial head n demarcates a frontier between two different structural domains (see also Harley and Noyer 1999;Marantz 1997;De Belder et al. 2014). The domain below categorial n is reserved for lexical-that is, noncompositional, semantically unpredictable-meaning, whereas the domain above n hosts functional projections contributing compositional-that is, non-idiosyncratic, predictable-meaning. Another (related) property that distinguishes "low" (i.e., below n) and "high" (i.e., above n) grammatical formatives concerns productivity. Low grammatical formatives are not morphologically productive-that is, they apply only to a closed set of roots-whereas high grammatical formatives are.
Starting with the latter property-morphological productivity-I conclude on the basis of the examples in (37)  The impossibility of the examples in (37a -c ) shows that the CS-pattern in (31) is a non-productive one.
Let's next turn to the first property: non-compositional meaning. Although one may still recognize some of naam's original meaning (i.e., 'name') in namens, it is quite clear that its meaning has become more abstract, as is also clear from its English translation: 'on behalf of.' Another illustration of a 'root-en-s' form, whose original meaning (namely: 'way') is no longer present, is the form wegens (way-en-s 'because of'), which designates the cause or reason of something, as in wegens de storm 'because of the storm'.
The question as to what the element -en contributes to this more abstract meaning obviously arises. I propose that, in present-day Dutch, -en is a derivational morpheme with a classifier-like function. Specifically, I assume it has a dividing function in the sense of Borer (2005:109-112); it portions out the mass denoted by the root and turns it into a unit. 29 As such, -en in (31) has essentially the same role as the (singular) indefinite article een 'a' in the noun phrase een naam 'a name.' Interestingly, and maybe not surprisingly, the bound morphemic -en in namens has the same phonological form as the indefinite article een 'a' in a noun phrase like een naam (a name), namely /@n/. I propose that -en in namens is a "low" (i.e., derivational) classifier, while the indefinite article een is a "high" classifier, that is, a classifier representing a functional category within the extended nominal projection. I take this "high" classifier to be located right above the nP-layer: . 30 Having argued that -en in namens is a classifying bound morpheme that turns a root into a unit, let me give some evidence in support of this unit interpretation of the bound morpheme -en. Consider the following examples: 29 The underlying assumption here is that the mass reading is the default reading of a root; see Borer (2005). 30 Note the parallel with diminutives. As De Belder et al. (2014) have argued, a distinction should be made between "high" diminutives, which occupy a functional head, and "low" diminutives, which are derivational morphemes occupying the head position of LexP. Interestingly, Wiltschko's (2005) interpretation of diminutives as being classifiers, leads us to conclude that the distinction between "low" classifiers and "high" classifiers is a more general phenomenon. In (38a,b,c), the bound morpheme -en represents a unit of the class "spatial location." In (38a) we have a location defined in terms of a compass point, in (38b) a geographical location representing a country (Zweden, Polen) or a city (Mechelen, Groningen, Leiden), and in (38c) a location defined in terms of a spatial point: 'the back (side),' 'the front (side).' In (38d,e), -en represents a unit of the class "temporal location/time." The phrase na enen in (38d) can be paraphrased as na één uur (after one hour, 'after one o'clock') and na vieren as na vier uur (after four hour, 'after four o'clock'). 33 In (38e), -en represents a temporal unit of the type 'day,' as is also suggested by its English equivalent: yester-day.
A striking feature of the examples in (38) is, of course, the (obligatory) absence of -s after -en. The pattern na enen-s, for example, is impossible (compare with (38d)). I tentatively propose that this is due to the fact that -en in (38) is not a classifying derivational morpheme-so it is not the head of LexP-but rather an indefinite article that is used pronominally and substitutes for the entire functional projection ClP. Thus, na enen in (38d) has the following structure: 34 Having shown that forms such as namen (i.e., [naam + -en]) are non-productive (see (37)), and have a non-compositional meaning, I now turn to the derivation of the surface form namens, which must feature so-called adverbial -s. For this, let us return to the abstract representation in (36), which is repeated in (39a). Following De Belder et al. (2014), I assign the label LexP to the structural layer located in between the root and nP. This gives us the base structure in (39a). As indicated in (39b,c), I 31 The element mid occurs also in phrases such as (tot) mid januari '(until) mid January' and complex words, as in de midvoor van dit voetbalelftal 'the center-forward of this soccer team.' 32 That these forms have a composite structure is shown by the fact that the root can also be part of another complex form, as in een Zweed-s-e actrice (a Swede-s-INFL actress, 'a Swedish actress') and een Mechels-e herder (a Mechel-s-INFL shepherd, 'a Belgian Shepherd (Malinois) dog'). 33 The fact that -en co-occurs with één, as in na enen, shows that -en does not correspond to the Dutch plural marker -en, as in hoed-en (hat-s, 'hats'). This means that -en in na vieren is not a plural marker. 34 I leave a systematic investigation of the patterns in (38) (39c). With namen and n both being bound forms, we end up with an exocentric head-head configuration, whose subparts share the property of being bound elements. I propose that in a configuration like this, namen and n must externalize conjointly at PF. Their mutual dependence-both being bound elements-requires the two elements to materialize together, yielding a single lexical unit at the surface. I take externalization of n as -s to be a last resort operation: -s-support must take place in order to externalize the exocentric complex head.
Having given an analysis of the internal structure of nominal expressions such as namens, I can now give an account of the ill-formed examples in (32). As pointed out above, I assume that the pattern namens de regering represents a Construct State pattern, with namens being the Construct State nominal (the possessum) and de regering being the possessed nominal. As shown in (40a), I propose that de regering occupies the specifier position of a functional projection encoding possession (i.e., PosP). In line with Cinque (2005) and Shlonsky (2004), I take the surface pattern to be derived by means of phrasal movement of the possessum across the possessor, as in (40b). Specifically, nP (namens) raises to [Spec,DP], the edge position of the nominal expression, and moves on to [Spec,PP], which is a potential landing site for displaced material in Dutch (see van Riemsdijk 1978 Summarizing, I have argued that the adverbial -s that is part of apparent adpositions such as namens 'on behalf of,' is an affixal realization of the categorizing head n. Its appearance is induced by the bound-morphemic nature of the two subparts of the head-head amalgam, namely n and the bound form namen.

Adverbial -s in small nominal expressions
This section examines the patterns ondergronds (under+ground+s) and binnensmonds (inside-s-mouth-s). These patterns will be analyzed as adpositional structures that contain nominal expressions featuring adverbial -s. It will further be proposed that, when these patterns act as attributive prenominal modifiers, they do not behave like PPs but rather like adjectival phrases.

The pattern P+N+-s
This section examines the italicized pattern P+N+-s in (42) inside-mouth-s 'Jan was cursing silently.' As indicated by the small capitals, phonological stress falls on the nominal element at the end of the italicized string. In Sect. 2, I already noted that this corresponds to a phrasal stress pattern and not to a compound stress pattern. Thus, the italicized patterns in (42) are constituents with a phrasal syntax.
Additional support for the syntactic nature of the italicized patterns in (42) comes from pronominalization. As shown in (43), an R-pronoun can substitute for the nominal component of the sequence P+N+-s (e.g., kamers in (43a)). 36 This substitution is similar in nature to the one attested in the minimal pair binnen de kamer (inside 35 Observe that binnensmonds in (42c) also displays adverbial -s in the middle of the linguistic expression. In traditional grammars, this -s has been characterized as 'proleptic -s'; it anticipates the appearance of the final -s. The nature of this proleptic -s will be discussed in Sect. 4.3.2. 36 See also my earlier discussion of example (10)  Given the surface similarity between the italicized patterns in (42), on the one hand, and the prepositional phrases onder de grond (under the ground), bij de tijd (by the time) and binnen de mond (inside the mouth) in (44), on the other hand, it is tempting to analyze the italicized elements in (42) as PPs. 37 Notice that, just as in (42), the nominal element at the end of the prepositional structure in (44)  mouth 'There were ulcers inside the mouth.' A PP-analysis of the sequence P+N+-s is supported by a number of grammatical properties. Firstly, the sequence can combine with (measure/degree) modifiers that are found in prepositional contexts, as in ruim/enkele meters/ver on de grond (amply/several meters/far under the ground). Some illustrations, drawn from the internet, are given in (45) Thirdly, the sequence P+N+-s displays the distributional behavior of a PP. For example, PPs that function as modifiers within a noun phrase are typically postnominal. In this respect, they behave differently from attributive APs, which typically precede the noun. This contrast in distribution is clearly shown by an example like die [ AP vreselijke] ruzie [ PP binnen dat gezin] 'that terrible fight within that family.' As shown in (47), the sequence P+N+-s can also be postnominal. Thus, distributionally it behaves like a PP. Another distributional property that hints at the prepositional status of P+N+-s is the fact that it can occur in postverbal position (the so-called PP-over-V phenomenon ;Koster 1975). This is shown in (48). In Dutch, only PPs and clauses (CPs) can occur in postverbal (i.e., extraposed) position. On the basis of the three above-mentioned prepositional characteristics, I conclude that the sequence P+N+-s must be analyzed as a PP. This means that the examples in (42a-c), which instantiate this pattern, have the base structures in (49a-c) and the derived structures in (49a -c ). 38 As indicated, I take the root to head-move to the categorial head n. The question arises as to why n must surface phonologically. Importantly, n does not surface as -s when the complement of P is a full DP, as in onder de grond(*-s), lit.: under the ground(-s). I propose that the appearance of -s in linguistic expressions such as ondergronds relates to the nature of the root. Specifically, I take the root to be a bound root, that is: √ grond BM ('BM' = bound morpheme). In this respect, it differs from the root grond in onder de grond, which I take to be a free morpheme. This distinction between √ grond BM an free-morphemic √ grond receives support from their different behavior. Firstly, the pattern onder de grond is productive in the sense that any "noun" (i.e. n+root) that can represent the so-called Ground is possible in this configuration: onder de {tafel, boom, modder, arm, . . . }(under the table/tree/mud/arm). Notice now that the pattern ondergronds is non-productive. Many roots are excluded in this configuration: *onder {tafels, booms, modders, arms, . . . }. Secondly, the meaning of the pattern onder de grond is typically compositional (i.e., can be derived from its parts), while the meaning of patterns such as ondergronds is often not. As shown in (50) Having shown that there is support for the distinction between free-morphemic √ grond, on the one hand, and bound-morphemic √ grond BM , on the other hand, I return to the question about the appearance of -s. In onder de grond(*-s), -s does not surface as a realization of n because √ grond, the host of affixal n, is a free root. In ondergronds, on the other hand, the root √ grond BM raises to n, yielding an amalgam whose components are bound elements: [ √ grond BM + n aff ]. It is in this exocentric head-head configuration that the two components must surface together at PF. Their mutual dependence requires the two elements to materialize together, yielding a single lexical unit at the surface: gronds.
Having provided an analysis of the pattern ondergronds, I would like to discuss two other phenomena relating to this construction, namely (i) the appearance of a proleptic -s in expressions such as binnensmonds (see (42c)), and (ii) the appearance of ondergronds as an attributive adjectival modifier, as in een ondergrondse kamer (an underground-s-INFL room, 'a subterranean room'). I'll start my discussion with proleptic -s.

Proleptic -s
The attentive reader will have noticed that in expressions such as binnensmonds (42c), binnenskamers (43a), binnenshuis (43b) and buitenshuis (45c), there is an -s interspersed between the preposition and noun+s. Traditionally, this interspersed -s has been characterized as 'proleptic -s', since it anticipates the occurrence of the final -s, as in binnensmonds. Since -s typically surfaces in nominal environments, one would expect there be some nominal element present. In order to find this nominal element, we need to look into the "preposition." Given the fact that proleptic -s is absent in expressions such as onder(*-s)gronds (42a) and bij(*-s)tijds (42b), it does not seem implausible to assume that there is some kind of nominal element present within "prepositions" such as binnen, boven, and buiten. This obviously implies that these apparent prepositions have a composite structure. I propose that they can be decomposed into three parts (see also Royen 1947Royen -1954 and Aboh 2010 for Dutch, and Pretorius 2017 for Afrikaans).
(51) a. binnen = be-+ in + -en ('inside') b. buiten = be-+ uit + -en ('outside') The element be-has a prepositional flavor in the sense that it designates spatial information (specifically, location). This is clear, for example, from the slightly archaic patterns in (52) Presumably, the P-like element be-in (52) is the same element as in (53b), where beis a prefix on the verb laden and alternates with the locative preposition op in (53a). Following Hoekstra (1988), I take be-to be an affixal preposition designating Place, which has been incorporated into the verb (laadde). Let us next turn to the other two components in (51): -in/-uit/-ov and -en. I propose that -in/-uit/-ov are roots having a spatial meaning. 40 I assume that the element -en is the same element as in namens (name-en-s, 'on behalf of'); see Sect. 4.2. This means that -en is a "low" classifier-like element that heads the projection LexP, which is located in between the root and the categorial projection nP. Schematically: Head movement of √ in and adjunction to -en yields the amalgam [ √ in+-en]. Recall from Sect. 4.2 that -en's function was to turn a root into a unit. Taking this perspective, the amalgams -innen, -uiten, and -oven are complex heads carrying the meaning 'inner side,' 'outer side,' and 'upper side,' respectively. As is clear from their meaning description, these nouns designate an axial part of an object, such as the (inner/outer) side or top of that object (Jackendoff 1996:14). Importantly, forms such as -innen do not occur as independent nouns; they must co-occur with the affixal preposition be-. In what follows, I will show how the bound-morphemic preposition be-and the bound-morphemic amalgam -innen "come together" and form the unit binnen (be+innen). Furthermore, I will provide an account of the double appearance of adverbial -s in a linguistic expression like binnensmonds.
Starting point of my analysis is Svenonius's (2006) proposal that certain adpositional projections contain a functional layer designating an axial part of an object, e.g., its top, front, or sides. Svenonius, for example, proposes that the English adpositional expression inside of the house has the structure in (55), where in designates the location, side the axial part, and (of ) the house the so-called GROUND.

(55) [ PlaceP [ Place in] [ AxPartP [ AxPart side] [ KP [ K of] [ DP the house]]]]
Building on this structural analysis, I propose that an expression like binnensmonds starts out from the base structure in (56), which has the possessive meaning 'in mouth's inner-side': According to this analysis, a possessive relationship is encoded in the lower part of the structure. The lower nP (complement of Pos) represents the possessum ('inner side'), the higher nP in [Spec,PosP] the possessor. In informal terms, the possessive relationship can be paraphrased as 'mouth's inner side.' Let us next consider the derivation of the surface pattern binnensmonds, starting with the possessum -innens. I propose that innens has exactly the same derivation as namens ('on behalf of') in Sect. 4.2. This means the following: the root in in (56)  and n being bound elements, we end up with an exocentric head-head configuration, the subparts of which share the property of being bound forms. I propose that in a configuration like this, the two heads must externalize jointly at PF. For n this means that it spells out as -s, a last resort operation. Let's next turn to the possessor monds in (56). I assume that monds has the same derivation as gronds (ground-s) in linguistic expressions such as bovengronds (aboveground-s); see (49a ). This means that mond is a bound root, that is: √ mond BM . This bound-morphemic root raises to n, yielding an amalgam whose components are bound elements: [ √ mond BM + n aff ]. Their mutual dependence requires the two elements to materialize together at PF, yielding a single lexical unit at the surface: monds.
Having accounted for the formation of the nPs innens and monds, let us finally consider the derivation of the word order pattern. Taking (56) to be the base structure, I assume that the nP innens undergoes phrasal movement from the complement position of Pos to the Specifier position of AxPart, located in the adpositional "middle field." This movement step, depicted in (57), yields the syntactic structure that surfaces as binnensmonds (i.e., be-[innens]-monds).

(57) [ PP be-[ AxPartP [[ Lex [ √ in]-en]+n (= -s)] m [ AxPart [Pos k +AxPart] [ PosP [ nP monds] [ Pos t k t m ]]]]]
It should be noted that the nP innen crosses another nP on its way to [Spec,Ax-PartP], namely monds in [Spec,PosP]. This looks like a violation of locality (Rizzi's (1990) Relativized Minimality or Chomsky's (1995) Minimal Link Condition); the nP monds is closer to [Spec,AxPartP] than is the nP innen, and should therefore be the target of movement. However, following Chomsky's (1993) locality theory in terms of equidistance, the nP innen may cross the higher nP monds as long as the two nPs are equally far away from innen's extraction site. Under Chomsky's assumptions, this situation is obtained by the application of domain-extending head movement, in casu Pos-to-AxPart, as represented in (57). This operation creates a minimal domain containing both the moved nP innen and the possessor-nP monds.

Ondergronds and binnensmonds as attributive adjectival modifiers
So far, I have given evidence in support of the PP-like status of linguistic expressions such as ondergronds and binnensmonds; see (42). It should be pointed out, however, that these same expressions-at least, superficially the same-also display adjectival behavior. For example, they can be used as attributive adjectival modifiers, as shown in (58). Their adjectival nature is clear from the fact that they carry attributiveadjectival inflection (-e), and occur in prenominal position, the characteristic position of attributive adjectives; compare with (47), where binnensmonds and binnenskamers are postnominal and lack an adjectival inflection. The question, obviously, arises how to analyze these adposition-like expressions that display adjectival behavior. One option would be to say that there is an adjectival categorizer a that selects a PP, as in (61) An alternative analysis for adjectival modifiers such as ondergronds and binnensmonds would be to say that their adjectival behavior is due to the fact that grond 41 I would like to thank a reviewer for this suggestion.
(ground), mond (mouth) et cetera are roots that can be turned into an adjective by combining with the categorizing head a, as in (62). Importantly, and as will be discussed more elaborately in Sect. 6, this categorizing head can be realized as -s, just like n.
The adjectival behavior of root+-s amalgams is more common, as is clear from examples such as een zomer-s-e dag (a summer-s-INFL day, 'a summery day'), zijn slaafse gedrag (his slave-s-INFL behavior, 'his slavish behavior,' and een Deense taalkundige (a Dane-s-INFL linguist, 'A Danish linguist'). A question triggered by the analysis in (62), however, concerns the fact that an adposition takes an adjectival phrase (aP) as its complement. In general, adpositions typically combine with noun phrases, and not with adjectival phrases. It should be noted, though, that patterns can be found in Dutch, where an attributive adjective follows an adposition, as in een ver(re) van eenvoudige oplossing (a far from simple solution); see van Riemsdijk (2001). Importantly, this pattern features a modifier (ver(re)) before the adposition van. As was shown in (59), such modifiers are impossible in patterns such as ondergronds, which suggests that patterns such as ondergronds(-e) should be distinguished from the pattern verre van eenvoudig (-e). Notice that, also for the analysis in (62), it is not entirely clear why modification of the adpositional phrase is blocked. I finish this section with a third, quite tentative proposal regarding adjectival patterns such as ondergronds and binnensmonds. Suppose these attributive modifiers do not represent a P+complement structure but instead constitute exocentric phrasal structures in which onder and binnens act as phrasal modifiers adjoined to an attributive aP, which itself acts as an attributive modifier in (58)  I assume that gronds and monds constitute Root BM +a amalgams that express a relational meaning. They do not denote a property of the noun they modify, as oude 'old' does in een oude grafkamer 'an old tomb,' but express a relation between two entities. In een ondergrondse grafkamer in (59b), this relation is between grafkamer (the modifiee) and gronds (the modifier), and in dat binnensmondse gemompel in (58a) between gemompel (the modifiee) and monds (the modifier). It is the spatial modifier (onder and binnens) adjoined to aP that specifies the nature of the relationship, viz., a spatial relationship. It does not seem unlikely that we have the same structure in Dutch expressions such as extra-parlementair 'extra-parliamentary, outside of parliament' and intra-musculair 'intra-muscular, inside of muscles,' which feature the Latin-based elements extra 'outside' and intra 'inside,' which have spatial meaning. I will leave an in-depth analysis of the patterns in (58) for future research.

Adverbial -s: A cross-adverbial phenomenon
So far, I have presented three case studies on adverbial -s, which I analyzed as a manifestation of n. It was proposed that this categorizing node surfaces as -s in three structural configurations: In the pattern zachtjes (3.1), n forms an amalgam with the silent root WAY and surfaces as -s: [ √ WAY+n (= -s)]. In the pattern namens (3.2), n forms an amalgam with the complex head [[ √ na(a)m]-en], which features the "low" affix -en and constitutes a bound form that cannot occur on its own. In ondergronds, n forms an amalgam with a bound root: [ √ grond BM + n (= -s)] (4.3.1). The purpose of Sect. 5 is to show that adverbial -s, hidden though it may be, is a quite common phenomenon in Dutch morphosyntax, and that the structural configurations discussed in Sect. 3 are also attested in other types of nominal expressions featuring adverbial -s.

Temporal expressions with adverbial -s
Dutch has quite a number of temporal adverbs featuring adverbial -s at the end, such as eens (one-s 'once/one time'), opeens (at-one-s 'at once'), ineens (in-one-s 'at once'), soms (some-s 'sometimes'), and straks (soon-s 'soon'). There are also temporal expressions that have a proleptic adverbial -s at the beginning of the temporal expression besides having one at the end, as in 's avonds (-s-evening-s 'in the evening'), 's ochtends (-s-morning-s 'in the morning'), 's nachts (-s-night-s 'at night') and 's middags (-s-afternoon-s 'in the afternoon'). 42 Let's start our discussion with the patterns eens and op/in+eens, in which adverbial -s is attached to the numeral een 'one' (pronounced /e:n/). In op/in+eens, a preposition precedes the sequence een+-s. Phonological stress does not fall on the preposition but on the numeral: inEENs, opEENs. This stress pattern hints at the presence of a phrasal structure rather than a complex word structure, specifically, a prepositional structure: [ PP in/op [ XP eens]]. 43 The fact that these adverbs can be paraphrased by phrases featuring an overt noun designating (a point in) time-namely the noun keer 'time'-hints at the presence of syntactic structure: in één keer (in one time, 'at once'), op 'n keer (at a time, 'once'). Importantly, when the noun keer is present, -s cannot appear: in een(*-s) keer(*-s). I interpret the contrast between ineen*(-s) and in een(*-s) keer(*-s) as follows: in the former pattern, the categorial head n combines with a silent root (TIME), as in (64). 44 I assume that in this complex head configuration featuring a silent root-that is, [ √ TIME+n]-the categorizing head n externalizes as a last resort. Since the root does not get spelled out, n must surface overtly, the underlying assumption being that in a head-head amalgam at least one of the elements must externalize. Dutch has the weak adverbs 'ns (/@ns/) and 's (/@s/), which carry the same meaning: 'once/one time.' I assume that the sequence 'ns has a composite structure consisting of the elements -en (i.e, /@n/) and -s. I propose that /@n/ is the classifying derivational morpheme that was identified in expressions such as namens ('on behalf of') and binnensmonds ('inarticulately'). This means that the temporal adverb 'ns has the following base structure, where -en instantiates the  -en]; that is, the latter element cannot occur on its own. As for the temporal adverb 's, I assume it lacks the LexP-layer and has the following derived structure: . 45 As we will see below, this small temporal n also occurs in temporal expressions featuring proleptic -s. The approach taken so far can be extended to temporal adverbs such as soms 'sometimes' and straks 'soon'. I propose that they both have a composite structure featuring a silent root designating 'time.' Soms has the structure in (65a), with som realizing the Num-head, while straks has the structure in (65b), with strak being an adjectival modifier of the silent temporal nP. 46 (65) a Let's, finally, consider temporal expressions such as 's avonds (-s-evening-s, 'in the evening'), which display an adverbial -s both at the beginning-so-called proleptic -s-and at the end of the temporal expression. Historically, the initial -s is a reduced form of the element des, which can still be found in archaic temporal expressions such as des zomers (the-GEN summer-GEN 'in the summer'); see Royen (1947Royen ( -1954 46 In Standard Dutch, som is found in combination with the suffix -ig, as in sommige mensen, some+ig+INFL persons, 'some people'). Certain Dutch dialects use som in those contexts, as in Kempenland Dutch som minse 'some people' and som joore 'some years'; see De Bont (1958).
Construct State (CS) possessive noun phrases. In line with my analysis of namens (name+en+s, 'on behalf of') in Sect. 4.2., I take the word order 'possessum + possessor' to result from phrasal movement of the possessum to the edge (i.e., specifier) of DP. The possessum surfaces in a minimal way, namely as -s, which I take to be a realization of n. The possessor, on the contrary, features an overt root carrying -s: Evidence in support of the displacement analysis in (66) comes from the temporal expressions in (67), where the adjectival modifiers vroeg 'early' and laat 'late' follow 's ochtends.

The measure expression ietsje(s)
This section discusses the (optional) appearance of adverbial -s on the measure phrase ietsje(s), which can be decomposed into three parts: the element iets 'somewhat,' the 47 Compare gronds in bovengronds in Sect. 4.3.1. 48 In fn. 22, I noted that, in this article, I would not address the question as to whether expressions featuring adverbial -s are embedded within a PP headed by a silent P. It was shown that the manner expression zachtjes could not be extraposed, suggesting that it is a nominal expression, while a temporal expression like die dag 'that day' could be extraposed, suggesting that the latter is a PP. Notice that, superficially, P-less temporal expression such as soms, strakjes, and 's avonds can occur in extraposed (= postverbal) position; see (i). This suggests that these temporal expressions featuring -s are embedded within a PP. I leave a systematic study of this issue for future investigation. In order to come to an analysis of the pattern ietsjes, we need to examine more carefully the individual components of ietsje(s). 50 As a first step, I make a brief remark about the historical origin of iets, because it can help us in discovering its inner structure in present-day Dutch. Historically, iets derives from the sequence ie-('ever/sometime') + wicht ('thing') + wes ('what GENITIVE '); see Van der Sijs (2010). 51 The sequence ie-wicht got reduced to the form iet, and wes to the form -s, together yielding iets. 52 I assume that, in present-day Dutch, iets still has a composite structure: iet+-s. I take the component iet to be a bound root carrying the meaning 'thing.' In present-day Dutch, this bare form iet does not occur as a free morpheme but is still found in certain fixed expressions, as, for example, in the fixed coordinate pattern iet of wat (something or what, 'somewhat, a bit'). 53 I propose that the bound root √ iet raises to n, yielding the amalgam [ √ iet+n], as depicted in (70): 54 49 Iets can also have a referential reading, that is 'something,' as in Jan vergat iets (Jan forgot something). Interestingly, this referential iets cannot combine with a diminutive morpheme: Jan vergat iets(*je); 'Jan forgot something (small).' I tentatively propose that the referential reading of iets is obtained by moving it into [Spec,DP]; compare with (71b), where movement of iets to [Spec,QP] yields the quantificational/measure reading 'somewhat.' 50 Next to the measure expressions iets and ietsje(s) Dutch has the patterns ietswat and wat, which also have the meaning 'somewhat.' I leave the analysis of these patterns for future research, but would like to point out that wat, as opposed to iets, cannot combine with the diminutive -je: *wat-je (intended meaning: 'somewhat'). Possibly, the non-co-occurrence of wat and -je is due to the fact that they compete for the same syntactic position, namely Q(uantifier). The quantifier-like status of wat is clear from its appearance in noun phrases such as wat boeken (what books, 'some books'). 51 Note that, in the [+HUMAN] indefinite pronoun iemand 'someone', the counterpart of wicht 'thing' is still identifiable: ie-man-d (ie + 'man' + paragogic /t/, 'someone'). 52 The sequence iet+wes is still visible in the German form etwas ('something, somewhat'). 53 In certain present-day Dutch dialects, the -s-less form iet is still used as a free morpheme . 54 As a reviewer points out, the nominal status of iets is also supported by the contrast between Dutch niets and niet. The former means 'nothing' and is the negative counterpart of iets 'something,' the latter is the negative word (meaning 'not') that is used for expressing sentential negation or constituent negation. I assume that niet is a functional category (say, a Neg-head) with a non-composite form, which means that the nominal component iet is not present in niet. Notice at this point that iet does not function as the (emphatic) affirmative counterpart of niet. For this, Dutch uses the word wel (emphatic affirmation). The element niets 'nothing,' on the contrary, does have a composite structure: n-iet-s, where the subpart iets has the structure in (70). I tentatively propose that the negative component n-in niets occupies the Q-head that selects nP. So we have the following structure: . Potential support for The components of this amalgam have an equal status in the sense that they are both bound morphemic (BM) elements. I assume that in this exocentric head-head configuration-[ √ BM + n affixal ]-the two components externalize jointly at PF. Their mutual dependence-both being bound elements-requires that the two elements materialize together, yielding a single lexical unit at the surface. I take externalization of n as -s to be a last resort operation: -s-support must take place in order to externalize the exocentric complex head.
Having provided an analysis of iets, let's next examine the alloforms ietsje(s) in (69). Following my analysis of diminutive -je in zachtjes (silent/soft-DIM-s, 'silently/softly'), as given in Sect. 4.1, I propose that -je is located in Q. With -je occupying Q, we end up with the base structure in (71a) for ietsje. I propose that the sequence ietsje is derived in two steps: (i) head movement of √ iet to n, as in (71a), followed by (ii) nP-movement to Spec,QP. The derived structure is given in (71b), where only the second movement step has been represented.
Let's now turn to ietsjes, which displays an adverbial -s at the end of the string. I propose that ietsje and ietsjes differ from each other only in the richness of Spellout of the movement chain. In ietsje (71), n materializes only within the head of the nP-movement chain, that is within the nP occupying [Spec,QP]. In ietsjes, on the contrary, n materializes both within the head of the nP-chain and within the foot of the nP-chain. This is depicted in (72), which is derivationally preceded by (71a): 55 The question, of course, arises as to why the categorizing head (optionally) surfaces twice in (72). Possibly, ietsjes is a slightly more emphatic variant of ietsje. Summarizing, I proposed in this section that the measure expression ietsjes is a linguistic expression with a composite structure. Both the interspersed -s and the final -s are manifestations of the categorizing node n. The presence of interspersed -s is induced by the bound-morphemic nature of the root iet. The final (optional) -s, was this analysis comes from the complementary distribution of diminutive -je, which I take to be in Q, and negative n-, as exemplified in (i): lang. tall 'I think he is somewhat too tall/in no way too tall.' 55 For multiple spell out of chain positions, see, among others, Nunes (2004). The multiple Spell-out approach in (72) could possibly be extended to partitive-genitival constructions of the type iets (erg) moois (something-s very beautiful-s, 'something (very) beautiful'), where the adjective is followed by -s. Rather than interpreting -s as being part of the adjective, one might analyze it as a minimal spell-out of the nP-copy, as depicted in (i). In (i), the nP iets raises across the attributive AP to [Spec,DP], the locus of referential information. analyzed as a manifestation of the trace/copy-specifically, its n-component-left behind after nP-movement.

The degree expression vliegens
This section examines the occurrence of adverbial -s in structural contexts like (73) In each of the examples, we seem to have a verb-like element (e.g., vliegens) that acts as a degree modifier of a gradable adjective (e.g., vlug). 56 An expression like vliegens vlug in (73a) has the paraphrase 'so fast that you are flying/fast as if you are flying.' This metaphorical reading corresponds to the degree reading 'very fast.' An important characteristic of the degree modifier is the presence of adverbial -s. This -s is obligatory: *vliegen vlug. 57 Before turning to an analysis of -s, I should point out that the sequences vliegens vlug and stervens koud are not compounds but phrasal constituents. Evidence for this comes from phonological stress. Under a neutral intonation, phonological stress falls on the second element, as is characteristic of phrasal stress. Thus, stervens KOUD, and not STERVENS koud. 58 Importantly, stervens (and also vliegens) can become phonologically more prominent when they are used emphatically, which, obviously, is possible with degree modifiers. This emphatic reading of the degree modifier often leads to a stretched pronunciation in which each syllable receives its own stress: STÉR-VÉNS koud.
As shown in (74a), the degree modifiers in (73) cannot co-occur with adjectival degree modifiers such as erg 'very' and vreselijk 'extremely,' which suggests that 56 Besides the degree adverbs in (73), in which -s is attached to a verb-like element, Dutch has degree adverbs of the following type, in which adverbial -s is attached to a noun-like element: honds brutaal (dogs impertinent 'very impertinent'), doods bang (dead-s afraid 'deadly afraid'), bliksems goed (lightning-s well 'damn/very well'). I will leave these patterns for future research. Analyzing these patterns along the lines of (onder)gronds ((under)ground-s) in Sect. 4.3.1 seems to be a plausible way to go. 57 Certain Dutch dialects permit adverbial -s on manner-adverbials. Example (i) illustrates this for Katwijk Dutch (Overdiep 1940). Present-day Standard Dutch uses a present participial form in those contexts: lopend. I assume that forms such as loopes in (i) receive the same analysis as vliegens en stervens in (73b); see below in the main text. 58 Examples of compounds, with phonological stress on the infinitival form, are the following: het STER-VENSgevaar (the die-s-danger, 'the danger of dying'), de STERVENSduur (the die-s-duration, 'the length of the dying process'). Compare these forms with patterns featuring the degree element stervens: stervens GEVAARLIJK (die-s dangerous 'very dangerous'), and stervens DUUR (die-s fast 'very fast'). they compete for the same structural position. I take this position to be [Spec,QP] of the extended adjectival projection. As shown by (74b), vliegens can be preceded by the degree word zo 'so,' which I take to be the head of DegP, the layer on top of (adjectival) QP; see Corver 1997. 59 (74) a Let's now address the question as to why adverbial -s must appear in (73). For answering this question, we need to find out what the inner structure of vliegens is like. From a surface perspective, the subpart vliegen, to which adverbial -s is attached, looks like an infinitival form. As shown in (75) considers 'Jan will fly again tomorrow, since he likes flying.' As has been noted by traditional grammarians, however, forms like vliegens in (73) historically derive from present-participial forms (de Vooys 1967:257). In the case of vliegens, this is the form vliegend, which has -end as participial morphology. The final (dental) consonant of this participial morpheme disappeared, which led to the surface form vliegen.
Importantly, degree modifiers having an integral present-participial form -end are still possible in present-day Dutch. Some examples are given in (76). Note that the -s-variant is impossible here.  ';Sect. 4.3.2). Under this analysis, -en is a derivational morpheme with a classifier-like function, more specifically a dividing function in the sense of Borer (2005). I assume it portions out the mass denoted by the root vlieg and turns it into a unit, say 'flying.' This unit is a degree unit, that is, a point on the scale of degrees. It represents a degree that equals 'flying.' The derivational steps for vliegens are given in (77) and are similar to those for namens (see (39)) and (b)innens (see (57) vliegen and n both being bound elements, we end up with an exocentric head-head configuration, whose subparts are morphologically similar (i.e., bound-morphemic).
In a configuration like this, the two heads externalize jointly at PF. Their mutual dependence-both being bound elements-requires the two elements to materialize together, yielding a single lexical unit at the surface.

Adverbial -s as a manifestation of a
In Sects. 4 and 5 we have come across various nominal expressions featuring socalled adverbial -s. It was proposed that -s is a manifestation of the categorizing node n, and that it surfaces in specific structural configurations. In this section it will be shown that adverbial -s also appears in adjectival expressions. I will propose that -s can be a manifestation of the categorizing node a. 60 Importantly, and maybe not 60 I follow here Marantz (1997Marantz ( , 2000, who argues explicitly for the existence of the categorizing head a next to n and v; see also Lowenstamm (2014). Marantz claims that a encodes a "property" at the interpretive level. See, however, Fábregas and Marín (2017), who point out that the property reading is not unique to the category 'adjective.' Certain nouns (e.g., hue) and verbs (e.g., to exist) also seem to denote properties. On the basis of this observation but also the argument that adjectives do not display the full range of behavioral properties attested for n and v -e.g., the absence of light adjectives, periphrastic adjectives and semi-lexical adjectives- Mitrović and Panagiotidis (2020) propose that the categorizing head a does not exist. According to them, the only categorizing heads are the nominalizer n and the verbalizer v. What they call 'adjectives' are syntactically derived objects that take a two-layered structure consisting of an nP-layer and a higher vP-layer as their base. For reasons of space, I'll abstract away from an in-depth discussion of the important question as to whether a exists as a separate categorizing head or whether it can be reduced to the categorizing heads n and v. Similarly to what was observed earlier for nouns (see (15a)), the categorizing head does not surface when the adjective is "created" (derived) by head-moving the overt (free) root to the categorizing head a, which I take to be an affixal element. In line with what I have proposed for Root+n amalgams (see Sect. 3), I assume that, in [Root Free +a] amalgams, affixal a is adjoined to the raised root (see Chomsky 2015:12). With affixal a being embedded in the complex root, a is invisible from the outside. In this head-head configuration, a does not surface (i.e., externalize) as -s. The appearance of adverbial -s is blocked not only on positive forms such as zacht in (79)  was analyzed as a PP containing an nP, whereas prenominal ondergronds/binnensmonds was analyzed as an attributive (relational) aP to which P(P) was attached as spatial modifier. 61 Adverbial -s cannot appear either in between the root and the comparative morpheme: *zacht-s-er.
In both representations, affixal a is adjoined to the raised root. With affixal a being embedded in the complex root, a is invisible from the outside. Therefore, a does not surface (i.e., externalize) as -s. Although adverbial -s is typically not found on comparative forms such as zachter, there is a comparative form, however, that features adverbial -s, namely the manneradverbial form anders (other-s, 'differently'). Before giving a structural analysis of this form, I will make some basic observations about this pattern. First of all, anders exhibits properties of a comparative construction: it includes the bound comparative morpheme -(d)er (an-der-s) and it can co-occur with the comparative dan-phrase ('than'), as in (82) Secondly, -s is obligatorily present on anders. Thus, the -s-less form ander is excluded in (82). Notice that, in this respect, anders behaves differently from other comparative adjectival expressions featuring -er (see (80)). Importantly, the -s of anders is not an intrinsic part of the adverbial expression. Note, for example, that in its attributive use, as in (83) From the above data, it can be concluded that anders has a comparative meaning but does not display the full set of properties that we find with comparative adjectives that have a rule-based syntactic derivation, as in (81) above.
I propose that ander(s) is an adjectival expression whose comparative meaning is lexically specified. More specifically, I assume that -er is a "low" comparative derivational morpheme that heads the projection LexP, which is located in between the root-the bound morpheme √ an-and the categorial projection aP: 63 Importantly, the comparative form ander does not have a corresponding positive (*an) or superlative form (*het an-st). 64 This shows that anders is not a comparative form characterized by full productivity, and that it differs in this respect from a comparative form like zachter (80), which does have a corresponding positive (79) and superlative form (80). 65 The form anders is derived by raising the root to the "low" comparative morpheme, yielding the amalgam [[ √ an]-(d)er]. 66 Both components are bound forms: that is, they cannot occur on their own. I assume that, given the bound status of its components, the amalgam an+der also has a bound status. As was pointed out above, ander cannot occur on its own in a sentence like (82); -s-Support is required. 67  Let me briefly return to the properties exemplified in (82) and (84). I take the comparative meaning of anders to be defined within LexP, with -er being a derivational morpheme encoding comparison, and the dan-phrase (82) occupying [Spec,LexP], as in (87). The word order anders dan Piet results from the head movement steps depicted in (87).
syntactic level. This explains why patterns such as (84b) are possible, in which heel erg, a phrasal modifier that typically combines with a positive adjective (84a), cooccurs with anders. Notice also that comparative meaning and superlative meaning can be expressed syntactically in combination with the adjective anders: 69 Another ellipsis-like context in which a possibly surfaces as -s comes from certain Dutch dialects (spoken in Flanders) in which the wh-word (h)oe 'how' is followed by -s, as in (ia) from Aarschot Dutch (Pauwels 1958:392). Interestingly, -s does not surface when oe is followed by an overt adjective, as in (ib). In future research, I hope to examine these patterns in a more in-depth way. (90) represents a discussion between person A and person B. Person A claims that Jan loves Brussels sprouts, and Person B contradicts A's statement. In (90a), this is done by means of a copular construction consisting of the elements (i) da's, a contraction of the demonstrative pronoun (dat) and the copular verb (is), (ii) the negative adverb niet 'not', and (iii) the adjective waar. In (90a ), Person B's answer also corresponds to a copular construction, but now the negative adverb niet is followed by the small element -es. In (90b,b ), finally, Person B's answer is a fragment answer. In (90b), the fragment answer consists of the negative word niet and the adjective waar, while, in (90b ), it consists of niet and -es.
Before discussing the nature of -es in (90a ,b ), I should point out that we find the same patterns with the affirmative adverb wel. For example, in return to Person B's negative reply in (88) nietes?" not-es 'Jan finished with the question: "Is this true/the case or not true/the case?"' b. Jan sloot af met de vraag: "Is dit wel waar of niet waar?" Having given some support for an analysis in which niet (90) and wel (91) are modifiers of a phrasal constituent, I will now turn to an analysis of the patterns nietes and welles. I tentatively propose that they have the structure in (93): According to this analysis, niet and wel modify aP. The sequence -es represents aP. I propose that it instantiates the pronominalization strategy of ellipsis (Corver and van Koppen 2011). Specifically, I take -e (pronounced as /@/) to be a meaningless sound that fills the root position, possibly for reasons of emphasis. 70 As indicated, the root raises to a, yielding the complex head [[ √ -e] i +a]. So we end up with an exocentric head-head configuration, whose subparts share the property of being bound elements. In a configuration like this, the two heads must externalize conjointly at PF. As a result of their mutual dependence-both being bound elements-the two heads must materialize together, yielding a single lexical unit at the surface. I take externalization of a as -s to be a last resort operation: -s-Support must take place in order to externalize the exocentric complex head. 71 I will now turn to the third adjectival pattern featuring adverbial -s, namely the one in (78c), repeated here as (94) Traditionally, the (somewhat archaic) pattern blootshoofds is characterized as an absolute genitival construction (Royen 1947(Royen -1954Corver 2007). Other instances of this pattern are the following: goedsmoeds (good-s-heart/soul-s 'cheerfully'), blootsbeens (bare-s-leg-s 'with bare legs'), and blootsvoets (bare-s-foot-s 'barefoot'). Notice that these examples display two instances of so-called adverbial -s: one at the end of the expression and one interspersed between the adjective and the noun: blootshoofds. From a surface perspective, the pattern looks quite similar to binnensmonds (in-sides-mouth-s 'inarticulately'), which was analyzed as a prepositional structure in which P (i.e., be-) selects a Construct-State-like possessive noun phrase (see (57)). In what follows, I will develop a similar type of analysis for the pattern blootshoofds in (94), which I take to be an instance of an adjectival Construct State pattern. 72 I start with the observation that the absolute genitival construction in (94)  in. into 'Jan walked into the garden bare headed.' 70 Insertion of schwa is quite common in emphatic forms. See, for example, ik 'I' versus ikke (I-e 'I' (emphatic)), dat 'that' versus datte (that-e 'that' (emphatic)). 71 The sequence -es is possibly also found in interjective expressions such as hebbes! (have-es 'Gotcha!'), jakkes! (yuk-es 'Yuk!'). I tentatively propose that -es in these examples represents a small nP, with -e (schwa) as a minimal spell-out of the root and -s as a realization of n. For example, hebbes equals heb (= '(I) have)) + [ nP √ -e i +n (= -s) [t i ]]. The nP -es in hebbes possibly stands for the object (e.g., an insect or ball) that the speaker caught. The nP -es, as in jakkes, possibly, refers to the object that is evaluated (expressively) by jak. 72 See Hazout (2000) and Siloni (2002) for discussion of adjectival Construct States (CS) in Modern Hebrew. An example, drawn from Siloni (2002), is given in (i): (i) yalda girl.FSG yefat beautiful.FSG mar'e look.MSG 'a beautiful looking girl' As noted by Siloni (2002), this adjectival CS is limited to inalienable nouns (e.g., mar'e), and is typically (but not exclusively) found with body parts. Notice the parallel with the Dutch pattern blootshoofds in (94), where hoofd (head) is the inalienable possessum noun that belongs to Jan, the subject of the clause.
The two absolute constructions have a similar interpretation. They both mean: 'with the head (being) bare.' Thus, an adjective (bloot) is predicated over an External Argument ((het) hoofd). Besides this similarity of meaning, there are also a number of striking differences between the two constructions. First of all, there is a difference in word order: The absolute met construction displays a straight order (EA + predicate), whereas the genitival absolute construction exhibits an inverted order (predicate + EA). Secondly, the absolute met construction starts with an overt P (met), whereas the absolute genitival construction lacks an overt P. Finally, the two constructions differ from each other as regards the amount of structure that can be added to the adjective (bloot) and the noun (hoofd). In the absolute genitival construction, bloot and hoofd must be absolutely bare. For example, no degree word (96a) or comparative morphology (96b) can be added to bloot, nor is it possible to have a determiner preceding the noun (96c In the absolute met construction, both bloot and hoofd can be modified or specified, as, for example, in met het hoofd nog bloter (with the head even barer 'with the head even more bare'), and met het hoofd iets te bloot (with the head somewhat too bare 'with the head a little too bare').
In view of the similarity of meaning -that is, both constructions encode a predication relationship between a predicate (bloot) and an EA ((het) hoofd)-I propose that the two absolute constructions start out from the same underlying configuration, namely a small clause structure. I assume this structure consists of a functional head (Pred) that mediates between the predicate (the complement of PredP) and the EA (the specifier of PredP); see Bowers 1993;Den Dikken 2006. Example (97a) represents the absolute met construction in (95), and (97b) the absolute genitival construction in (94) 73 In Modern Hebrew, only simplex adjectives-that is, adjectives that are not formed by an adjectival formative (suffix)-can occur in adjectival constructs (Siloni 2002). Example (i), drawn from Siloni (2002), presents a minimal pair: the simplex form ge'e ('proud') allows a construct, but the suffixed form ga'avtan ('proud') does not. I assume that the inverted word order (Pred + EA) results from movement of aP to [Spec,PP]. 74 Both the EA (nP) and the predicate (aP) are minimal phrasal projections, in the sense that they consist solely of a categorizing head and a Root-complement.
In each projection, the root moves to the categorial head, creating the complex head [root+n/a]. I propose that -s-Support is needed because √ hoofd and √ bloot are bound roots. When the bound root raises to n/a, we get a complex head whose components are bound morphemes: √ bloot BM + a aff (= -s) and √ hoofd BM + n aff (= -s). In this exocentric head-head configuration, the two components must surface at PF. Because of their mutual dependence, the two elements must materialize together, yielding a single lexical unit at the surface: bloots and hoofds.
Summarizing, I have tried to show in this section that -s-Support also applies to the categorizing node a. On the basis of three case studies, it was shown that -s-Support in the adjectival domain is attested in the same types of structural configurations as -s-Support in the nominal domain. In short, the phenomenon of adverbial -s (reinterpreted here as -s-Support) is another illustration of the existence of cross-categorial symmetry. 75

Conclusion
This article examined the grammatical nature and distributional behavior of socalled adverbial -s in Dutch, a phenomenon that, to my knowledge, has received no or very little attention in modern theoretical linguistics. As for the grammatical nature of -s, I proposed that adverbial -s is a last resort affixal manifestation of the categorizing nodes n and a. On the basis of a large variety of adverbial expressions, I showed that -s-Support occurs in a restricted number of complex-headconfigurations, specifically: (i) [Root Silent +n/a (= -s)], (ii) [Root Bound +n/a (= -s)], and (iii) [[Root+Lex]+n/a (= -s)]. An important outcome of this study is that many linguistic expressions, traditionally identified as adverbs, are syntactic objects with an inner structure having nP/aP as a subcomponent. If my reinterpretation of adverbial -s as being an affixal realization of the categorizing nodes n and a is on the right track, it can be used as a diagnostic element for identifying other pieces of nominal/adjectival structure in the build of Dutch. Just for the sake of illustration, I have given in (98) a list of other "adverbs" featuring adverbial -s. Clearly, the phenomenon is quite common in Dutch and opens ways for further investigation.
ing the question about the existence of adverbs as a separate syntactic category; that is, do adverbs constitute a separate syntactic category (i.e., a lexical category) within the parts of speech (i.e., besides nouns, verbs, adjectives and (certain) adpositions) or can they be reduced to other syntactic categories? I hope to have shown on the basis of this case study on Dutch adverbial -s that it is worth trying to reduce so-called adverbs to other syntactic categories. For this reduction strategy we need to detect the inner structure of adverbs, including the linguistic atoms that hint at the presence of a particular type of structure. In this article we used the element -s as a sign for detecting nominal or adjectival structure within linguistic expressions that are traditionally categorized as adverbs. The question obviously arises to what extent this reductionist approach can be extended to other types of adverbs. For reasons of space, I confine myself to a single illustration: the quantifier (Q) al 'all.' This grammatical formative typically surfaces in nominal structures, as in al het water 'all the water,' and thus may be used as a means for detecting nominal structure within "adverbs." Some illustrations of Dutch adverbs featuring al are given in (99). I will leave an in-depth analysis of these patterns for future research.

(99)
altijd (all-time 'always'), aldus (all-thus 'thus/in that way'), aldaar (all-there 'there/at that place'), alvast (all-ready 'already'), alreeds (all-ready-s 'already'), alom (all-about 'everywhere'), aldoor (all-through 'all the time'), meestal (most-all 'mostly'), veelal (much-all 'often'), overal (over-all 'everywhere'), nogal (yet-all 'rather/somewhat') Another issue that I leave for future research is the question as to whether equivalents of adverbial -s can be found in other languages. 78 Recall from Sect. 3 that adverbial -s relates to genitival case, which, in line with Emonds (1985) and , I interpreted as an affixal noun, and more specifically as affixal n. Also here I confine myself to giving some illustrations of "adverbs" featuring an element that one might possibly want to analyze as a realization of the categorizing node n. My examples come from French and the pertinent element is the meaningless element de 'of,' which has been analyzed as a manifestation of genitival case. If 'genitival case' is affixal n, one might interpret the element de in the following French adverbs as a sign of the presence of nominal structure, that is: [ nP n (= de) + Root]. 79 Again, I leave an in-depth analysis of these patterns for future research.
syntactic structures. 80 If one follows Cinque (1999) in assuming that "adverbs" occupy the specifier positions of designated functional heads, one might expect there to be interactions (e.g. feature matching) between properties associated with the composite "adverb" and (properties of) the functional head in the clausal spine.
In conclusion, adverbial -s is a fascinating phenomenon, which is widespread but somewhat hidden in the build of Dutch. Its appearance is not arbitrary and unpredictable, but, rather, restricted and predictable. I hope to have shown that adverbial -s provides further support for the idea that categorizing heads are part of human language, and that so-called adverbs can be reduced to other parts of speech. Importantly, my investigation of adverbial -s has also shown that phenomena such as silent nouns and the construct state pattern are important elements of Dutch morphosyntax.