DP or not DP, that is the question. We provide an analysis of nominal complexes (NCs)1 by means of an NP analysis in Head-Driven Phrase Structure Grammar (HPSG). That is, we account for structures with prenominal and postnominal genitive phrases (PreGens and PostGens) such as (1a) and (1b) assuming that the head of the construction is the noun (NP analysis), not the determiner (DP analysis).

The NP analysis takes the noun as the element determining the internal structure as well as the distributional properties of the whole construction. The main approach to analyse NCs like (1) changed in the linguistic literature from an NP (cf. Chomsky 1970; Jackendoff 1977; Szabolcsi 1983) to a DP analysis at least since Abney’s dissertation (1987) becoming the most widespread account for NC structures in the 90s and the beginning of the 2000s (Abney 1986; Hellan 1986; Hudson 1987; Haider 1988; Bhatt 1990; Olsen 1991; Vater 1991; Szabolcsi 1994; Netter 1994; Adger 2003; Alexiadou et al. 2007; Heck & Müller 2006; Salzmann 2020; a.o.). This change towards DP was driven partly by theory-internal assumptions, for instance that all phrases have the same structure qua X-bar theory. Nevertheless, part of the linguistic community – mostly outside MGG3 approaches advocated the NP analysis (Baltin 1989; Pollard & Sag 1994: 359–376; Van Langendonck 1994; Sadler & Arnold 1994; Demske 2001: Section 4; Payne & Huddleston 2002; Van Eynde 2006; 2020b; Müller 2013: 88–88; 2020; Machicao y Priemer 2017; a.o.) and in the last decade, even in MGG approaches, the nP/NP analysis experienced a revival (Bruening 2009; 2020; Chomsky 2007; Georgi & Müller 2010; Chomsky et al. 2019; a.o.).

This article is organised as follows: Section 2 gives an outline of the development of the DP analysis showing how empirical evidence regarding the parallelism between sentences and NCs combined with theory-internal assumptions leads to the DP analysis. In Section 3, we discuss some DP approaches in German. We focus the discussion on (i) the element proposed as D head, (ii) the realisation of nominal arguments, and (iii) the structures proposed. As a kind of introduction to our analysis, we first give a short overview of HPSG (Section 4). In Section 5, we show how an NP approach can account for (i) the realisation of “external” and “internal”4 arguments of the head noun, (ii) theta role and case assignment, (iii) the asymmetry between external and internal arguments of a noun, (iv) the role of possessives within NCs, and finally (v) the semantic composition of N head and arguments. Furthermore, our analysis discusses (and allows for) complex and recursive structures in prenominal position which have been neglected in the literature on German PreGens. In Section 6, we compare our analysis with other possible NP approaches in HPSG.

Within Transformational Grammar, the parallel between nouns and verbs had already been pointed out by Chomsky (1965: 184–192). The underlying question is – taking the relation between head and arguments for granted – how to arrive from one form (e.g. destroy in (2a) to another (e.g. destruction in (2b)), and how the relation between these words must be accounted for.

At the beginning of the X-bar theory (cf. Chomsky 1970), the main focus is on the comparison between the structure of sentences and NCs. The idea underlying a parallel analysis of both is that a structural position correlates with function. Interestingly, the parallel made at this stage is between determiner and auxiliary verb (cf. (3) and (4)), with both functional elements occupying the specifier position of lexical phrases, NP or VP respectively (cf. Chomsky 1970: 210).5

In the further development of X-bar theory (cf. Jackendoff 1977: 36–37), the concepts of headedness and endocentricity are strengthened. In comparison to the earlier stage, it is assumed that a sentence (S) is a projection of the lexical head V (Vʹʹʹ in (5)) and not an exocentric structure (cf. (3)). This change leads to a welcomed parallel since the subject-of-relation can be modelled as a head-specifier configuration in NCs and sentences (Jackendoff 1977: 38–41; 165–166). That is, if John has the same interpretation in (5) and (6), then it has to be in a similar structural relation to the head of the dominating phrase in both cases.6 Hence, this analysis strengthens the idea of structural positions reflecting syntactic functions.

The further development of this approach strengthens the constraints imposed on structures, for instance w.r.t. binary structures and maximal projections (cf. Chomsky 1986: 2–4; Chomsky & Lasnik 1993: 527–533). Furthermore – and that is of significant relevance for the DP analysis – the functional categories COMP and INFL are introduced (cf. Chomsky 1981: 18–19; 1986: 4–6). INFL is a functional category replacing the auxiliary complex (for tense, agreement, and modality) on earlier stages of the theory (cf. (5)) and marking the relation between a subject and its predicate. COMP, on the other hand, is justified by the necessity of a position for complementizers (e.g. that) preceding the subject, as well as for wh-elements normally preceding a “subject-position” (Chomsky 1986: 5). COMP and INFL display the complete phrasal structure of lexical elements such as verbs or nouns (cf. (7) and (8)), following the principles stated by X-bar theory (Chomsky & Lasnik 1993: Section 3.2), in particular: (i) all phrases (of all categories) have the same structure, (ii) a head is not optional, and (iii) every phrase can have only one element identified as its head. A welcome effect of these principles is that sentences can be analysed as endocentric constructions and there is no need to analyse them as “defective categories” (Chomsky 1986: 5), e.g. as exocentric structures or C′ structures without phrasal level. Following these assumptions, taking functional elements (I in (7), C in (8), cf. Chomsky 1986: 73) as the head of sentential structures, and taking the observation that NCs and sentences show a parallel structural behaviour, logically leads to the analysis of the functional category D as head of NCs. This analysis is worked out in Abney (1987), but the basic idea can be found in the literature since the 1970s in some cases following the line of thought just presented (cf. Lyons 1977: 392; Vennemann & Harlow 1977; Brame 1982; Hellan 1986; Abney 1986; a.o.).7 The DP analysis was embraced by the linguistic community – not only in MGG approaches8 – and became the standard analysis for NCs since then.

As the development of the theoretical assumptions shows, it is closely linked to the development of framework internal assumptions, such as the axioms of X-bar theory, their assumption as a universal part of our language competence, and the strong correlation of position and function. In the case at hand, the parallel between sentences and NCs as well as between determiner–NC and subject–sentence relations have been recognised in more surface-oriented frameworks such as HPSG too (cf. Sag et al. 2003: 64). But one main difference between the two types of frameworks lies in the way of how pieces of linguistic information are modelled. While MGG approaches model single morphosyntactic or semantic functions (e.g. number, gender, tense, agreement, etc.) in different syntactic heads (cf. Alexiadou et al 2007: 227–235) inducing phrasal structures, surface-oriented frameworks deal with these functions as parts of observable linguistic objects such as words, affixes, etc. (cf. Section 4). In some cases, this competition for a “better approach”, i.e. accounting for more data with a more economic system, leads to a rethinking of the model used. For instance, the necessity of an IP-analysis for German clauses has been challenged by many researchers, also in the MGG tradition (Bayer & Kornfilt 1989; Haider 1993; a.o.). Thus, without an IP in German, the structural IP/DP parallel cannot be proposed, neither in German nor as part of UG.9 In Minimalism, since Hauser, Chomsky & Fitch (2002), the concept of UG has changed from a rich UG (comprising different modules such as X-bar theory, Case Principle, Theta Criterion, etc.) to a maximally simple UG reduced only to context-free recursion (cf. Richards 2015: 804–805).10 This theoretical change dissociated from X-bar theory, some assumptions about functional projections, etc., leads to a gradual rethinking towards an nP/NP analysis (cf. Section 3.2; Chomsky 2007: 25–26; Bruening 2009; 2020; Georgi & Müller 2010: 1–3; Chomsky et al. 2019: 22; a.o.). In the next section, the outline of the theoretical development just presented will focus on DP analyses in German, and especially on the realisation of PreGens.

In German, the prenominal position can be occupied by an element in the genitive – similar to English. PreGens can be syntactically simple or complex (9) and they are in complementary distribution with determiners (10). This complementarity suggests that determiners and PreGens compete for the same position, or at least have “something in common” syntactically (cf. Adger 2003: 257–258 for English). In the following sections, we discuss the realisation of the D head, case and theta role marking, and locality aspects of DP approaches for German NCs.

Our analysis of NCs is conducted in HPSG (cf. Pollard & Sag 1987; 1994). In contrast to the MGG analyses just discussed, HPSG is a model-theoretic approach (cf. Pullum & Scholz 2001; Richter 2007), belonging to the so-called constraint-based frameworks. In this spirit, linguistic theories are sets of constraints – formalised as attribute-value pairs – stating the necessary conditions on the structure of individual expressions (cf. Pullum & Scholz 2001: 19; Müller & Machicao y Priemer 2019: Section 4). HPSG models descriptions of linguistic objects (e.g. words, phrases, rules, etc.) by means of feature descriptions, which are formalised as attribute value matrices (AVM). AVMs consist of pairs of attributes (in small caps) and corresponding values of a certain type (in italics). For instance, in (27) the attribute NUMBER (NUM) has the value singular (sg). Values can be atomic or complex, the value of NUMBER in (27) is atomic (sg), while the value of INDEX (e.g. for the word man) in (28) is of type referential index (abbreviated as ref)25 and it is complex, i.e. the value itself is a feature description.

Every value, atomic or complex, and thus every feature structure, is of a certain type. Types are hierarchically ordered in the grammar from the most general types at the top to the most specific ones at the bottom. For instance, (29) shows the type hierarchy for number with its two subtypes singular and plural. While describing linguistic objects, we can use underspecified types to express generalisations. For example, it is possible to give a description of the linguistic object the as having the most unspecific value (i.e. number) for the attribute NUM, since it can be used in singular as well as in plural contexts. But in a particular utterance, the values must be maximally specific, e.g. sg in the child and pl in the children. The NUM value will be specified by means of an agreement constraint which states that the NUM values of a determiner and its corresponding noun must agree.

HPSG is a lexicalist framework, i.e. most of the relevant linguistic information is located in the lexicon, simplifying in that way the phrase structural component. Linguistic objects of all kinds (e.g. words, phrases, rules, etc.) are treated as signs in the spirit of Saussure (1916), i.e. form and meaning are always represented conjoined26 and always accessible. This is a major difference between HPSG and MGG. In the latter, syntax, semantics, and phonology are represented in different modules accessible only at different points of the derivation (cf. Chomsky 1995: 20–23; Richards 2015: 812; 830). In HPSG, the descriptions of linguistic objects are systematically stored in the (abstract) lexicon using the same formal mechanisms of description (i.e. feature descriptions) for all of them (cf. Müller 2019: 205–210). The organisation of the lexicon depicts the generalisations of all kinds of linguistic objects due to a type hierarchy reflecting different levels of abstraction (cf. Müller & Machicao y Priemer 2019: 330–331).

HPSG is surface-oriented and declarative, i.e. non-derivational. That is, structures are not derived from underlying structures, e.g. a passive from an active clause. Structures are analysed directly involving their observable elements in the positions they are. In general, concepts such as “movement” and “empty element” are avoided.27 The descriptions of linguistic objects given by HPSG focus on what can be observed without presupposing a universal syntactic structure connecting structural positions to grammatical functions. That is, the workflow in HPSG consists of (i) giving an adequate description of a phenomenon, (ii) formalising its generalisation, and (iii) – if possible – matching the generalisations in different languages arriving at a universal core of constraints (cf. Sag & Wasow 2011: 372; Müller 2014; 2015a). Hence, HPSG analyses do not resort to syntactic structures based on evidence of other languages than the one at issue.

As already mentioned, HPSG is a lexicalist framework, where lexicalist can be understood as in the following quote:

Following this kind of approach, we can give a description of Behandlung ‘treatment’ as in (30).

The AVM in (30) is the description of a stem (cf. type stem). It contains phonological (value of PHON(OLOGY)),28 and syntactic and semantic information (value of SYN(TAX)-SEM(ANTICS)). The value of SYNSEM is complex, it comprises different attributes (e.g. LOC(AL) and NON-LOC(AL) for local and non-local information).29 LOC is subdivided into syntactic information (in CAT(EGORY)) and semantic information (in CONT(ENT)). In CAT, there is an attribute called HEAD, whose value provides the information of the lexical head that is “projected”30 to the phrasal level by means of the Head Feature Principle (cf. Pollard & Sag 1994: 34). This information is particularly important for the distribution of the phrase, e.g. part of speech information.31 ARG(UMENT)-ST(RUCTURE) is a further syntactic feature. Its value is a list of synsem elements, the arguments of the linguistic object. This list does not represent the elements that the described expression syntactically requires, but it links the semantic valency of the predicate (in CONT) with the syntactic valency of the expression (SPR and COMPS – explained in Section 5.1). The elements of the ARG-ST list are ordered following the Accessibility Hierarchy (Keenan & Comrie 1977). That is, their order reflects their prominence w.r.t. independent linguistic phenomena such as passive, relative clauses, case, binding, extraction, etc. (cf. Manning & Sag 1998: 111; Koenig 1999: 29; a.o.). The semantic information is represented as the value of CONT. The value of IND(EX) is a complex value bearing the properties of the referential variable of the expression (cf. (28)). The value of REL(ATION)S is a list of elementary predications describing the meaning of the expression (cf. Copestake et al. 2005: 283). In (30), the list has only one element: the AVM for a treat-relation (treat-rel). This (Davidsonian) relation is a complex value with three attribute-value pairs: ARG0 has the referential the semantic arguments of the predicate, AG(ENT) and PAT(IENT) with their respective values and , are given as well. The indexed boxes in (30) are values (atomic or complex) shared with other parts of the structure, i.e. they state the notation for token identity. This representation of identity between (sub-)structures is called structure sharing and constitutes one of the most important descriptive devices in HPSG and other declarative frameworks (cf. Bildhauer 2014: 528–529; Müller 2019: 211–213; a.o.). In (30), the ARG0 value of treat-rel and the IND value of the stem are structure shared (cf. ). Furthermore, by the means of structure sharing, (30) states that the first element of the ARG-ST list32 of Behandlung is interpreted as agent () and the second as patient ().

In this section, we provide an analysis of NCs as NPs. In Section 5.1, we give a formalisation of the transfer of argument structure from verbal to nominal stems by means of lexical rules (LRs) reflecting the fact shown in (24), that event nominalisations are able to realise the arguments with structural case of their verbal counterparts. Since we are dealing with a lexicalist approach, the manipulation of argument structure must be addressed on a lexical level. In Section 5.2, we show how to account for the nominative-accusative to genitive case alternation from the verbal to the nominal domain. In addition, we outline a solution for the alternation between genitive NPs and von-PPs. In Section 5.3, we focus on the prenominal position. We discuss a parallel construction in German, the prenominal dative with possessive, that supports our analysis with an empty determiner for PreGens. Furthermore, we show that possessive elements (determiners, NPs in the genitive, von-PPs) interact with the argument structure of the noun being in complementary distribution with the external argument. This leads us to an analysis of possessives as arguments of the noun (cf. Barker 2012: 1114) accounting for the alternation by means of a LR. In Section 5.4, we show how the syntactic composition of the lexical elements explained in the previous sections leads us to an NP structure. In comparison to MGG approaches, functional properties of the structures are handled lexically, and not by means of phrasal structures. Therefore, we can account for the different types of constructions just presented respecting locality – as defined in Section 3.2 – and for the different interpretations of PreGens and PostGens without stipulating a quite general theta role for external arguments (cf. Kratzer 1996: 126–128). Our analysis is based on selectional localism (Sag 2012), i.e.

Therefore, the N head determines form and interpretation of the elements in its ARG-ST list, and the NC can be selected from outside according to the properties of the N head, not of D. Finally, in Section 5.5, we provide corpus evidence that the constructions with PreGens can be complex and recursive. Furthermore, we provide a semantic analysis accounting for the compositionality of these structures.

The data we account for have been discussed in (24). What has to be ruled out is a structure with a prenominal internal and a postnominal external argument (31a), as well as both arguments realised in prenominal (31b–c) or postnominal position (31d–e). Furthermore, for common nouns (e.g. Behandlung ‘treatment’), its realisation without a determiner must be ruled out (31f–h).

A different NP analysis in HPSG is based on the head-functor-phrase (cf. Van Eynde 1998; 2006; Allegranza 1998; 2007). In this approach (cf. (69)), the determiner (or PreGen) is not treated as a dependent of N, but as a functor taking the N head and marking it (Allegranza 2007: 260). This analysis is reminiscent of the NC treatment in Categorial Grammar, where the determiner (a functor) selects N licensing an NP/DP (cf. Vennemann & Harlow 1977; Bouma 1988; Steedman 1989; 2000).

The functor analysis works with the features MARK(ING) and SEL(ECT). The MARK value of a lexical item reflects its degree of saturation, i.e. unmarked, if it needs a determiner to form a complete phrase (e.g. singular common nouns, cf. in (69)), and marked, if it already has a determiner ( in (69)) or is fully saturated by itself (e.g. mass nouns) (Van Eynde 2020a: 7–8; 2006: 166–170). The SEL attribute is similar to SPEC (cf. Section 5.3.3). Its value constrains the type of head a functor combines with. For instance, the functors Jacobs or die in (69) select an unmarked N′ () licensing an NP. Since the head-functor-phrase is a subtype of headed-phrase, the head value of the head daughter (cf. ) is projected (cf. Footnote 30)66 Furthermore, the Marking Principle states that the MARK value of a head-complement phrase (cf. ) is shared with the head daughter, but for head-functor phrases (cf. , ) it is shared with the functor (Van Eynde 2006: 166; 2020a: 10; based on Pollard & Sag 1994: 400).

This approach has the following advantages (Van Eynde 2020b: xiv): First, it simplifies the phrase structural component accounting for head-adjunct and head-specifier phrases only with the head-functor-phrase. Second, it analyses NCs as NPs reflecting the pro NP arguments given here. Third, it eliminates the functional parts of speech (determiner, numeral, auxiliary, etc.), e.g. determiners are defined by means of their MARK and SEL values (Van Eynde 2006: 164). That is, in (69), Jacobs and die are fully saturated, marked “nouns” selecting an unmarked N′.

On the other hand, the prenominal position in German poses several challenges for the functor analysis. For instance, assuming the combination of argumental PreGen and noun licensed by the head-functor-phrase, N is not selecting the functor, hence it cannot determine the PreGen’s theta role (cf. (24b), (24c), (26)), and since functors cannot modify the valency of the head (Abeillé et al. 2004: 23), the PreGen cannot be taken to be an argument of N. Furthermore, in the functor analysis, possessive determiners provide the possessive semantics (Van Eynde 2020a: 12; see also Ginzburg & Sag 2000: 399): the functor selects N determining its referent as possessed. But possessive determiners and PreGens have a double life (cf. Section 5.3.2). They can be interpreted with different theta roles depending on the noun or with a possessive role (see LR (51)). To account for that (cf. (49a)), homonymous determiners would have to be proposed. Moreover, it would not be clear how a prenominal possessive can be interpreted as a patient, but only if the agent or possessor are not realised postnominally (cf. (49b)). To account for these data, the head would need access to the functor, leading to a double selection as proposed here (N selects DetP through SPR, Det selects N through SPEC), or to non-local selection, i.e. the functor would have to see which nominal arguments have already been realised (cf. (20) and its discussion).

A further problem of this account concerns the quantification of N. To avoid empty determiners, the functor analysis proposes that plural and mass nouns, i.e. elements that can appear without (overt) determiners (70a), have a lexically inherent (existential) quantifier (Allegranza 1998: 99–103). But since plural and mass nouns can combine with a determiner, this account is forced to delete the inherent quantifier when an overt determiner appears (70b) (cf. Allegranza 1998: 104 for technical details).67 Besides difficulties w.r.t. compositionality arising by deleting semantic material through syntactic combination (cf. Copestake et al. 2001; Bender et al. 2015), this account would also have problems with nouns in predicative structures (70c), where nouns are generally assumed to denote predicates and not entities or quantified elements.68 That is, the quantifier must be deleted as well, but not through addition of a different determiner as proposed in Allegranza (1998: 104).

Summarising, the functor approach provides a simplification of the grammatical component removing empty determiners, making the part of speech classification more parsimonious, and simplifying the phrase structural component.69 Nevertheless, the economical advantage of this approach decreases with the difficulties it encounters: different deletion rules for quantifiers must be added at a syntactic level, several possessive determiners with different theta roles must be included to the lexicon, pre and postnominal asymmetries are not accounted for, etc. In our approach, we handle these difficulties straightforwardly. The price we have to pay in some cases is a phonetically empty determiner, which is in any case empirically justified on the basis of the need for a quantifier.70 Furthermore, our (specifier) approach reflects the parallelism between sentences and NCs (2) since the DetP is treated as part of the valency of N, in contrast to the functor approach.

Another possibility to account for NPs in HPSG, proposed to us by a reviewer, is to analyse PreGens as raised arguments by means of weak heads. Weak heads are lexical heads introduced to deal e.g. with the non-prepositional uses of de in French, i.e. de-N′ combinations that do not behave as PPs, but as NPs, such as in beaucoup de livres ‘a lot of books’ (cf. Abeillé et al. 2004: 9; 2006: 156). Adapting this concept to NC structures, determiners would be analysed as weak heads.

In (71), the weak head would have to be an empty element71 that combines with a noun by means of the head-complement-phrase, so no further phrasal constraints (e.g. head-marker or head-functor-phrase) have to be added. Similar to the functor approach, weak heads use the MARK feature and the Marking Principle. The weak head selects for an unmarked N′ (cf. and ) with an empty COMPS list and projects its own MARK value, i.e. marked (). The peculiarity of weak heads is that they adopt the HEAD (), SPR (), and CONT values of their complements (). Therefore, when a weak head selects for a noun, the projected phrase is an NP. Furthermore, if the noun selected for a specifier, after the combination with the weak head, the new phrase selects for the same specifier () rendering a raising analysis of PreGens, i.e. weak heads are – per definition – subject raisers (Abeillé et al. 2006: 156).72 In the case of NCs with PreGens, the weak head has to be realised by an empty element, because weak heads are lexical heads (Abeillé et al 2006: 156), i.e. it is not possible to assume that (complex) prenominal genitive phrases are weak heads. Similar to the empty Det in (55), it would have its own semantic contribution and combine it with the semantic contribution of N.

But a weak head analysis does not work for NCs without changes in the central concept of weak heads. First, XARG is needed and a weak head has to raise its value, and not the element in SPR. Assuming that a noun has its PreGen in SPR, would lead to PreGen-noun structures (licensed through head-specifier-phrase) in which the noun is semantically not determined. Second, in the case of PreDats, the weak head cannot just raise the specifier of the noun, but it has to assign dative to it. Third, w.r.t. possessive determiners, the weak head itself needs an XARG value (cf. (56) and its discussion). A further (minor) problem concerns the parallel between NCs and sentences, since subjects are not expected to take the verb as a syntactic complement. A major problem of a raising analysis in general concerns constituency, since it implies a constituent structure in which the determiner combines with the noun, and the PreGen/PreDat combines with the resulting phrase (cf. (18)). As (72) shows, regardless of their complexity, the prenominal elements can occupy the same position and have a complementary distribution (10). This strongly suggests that the prenominal complex is a constituent (cf. Kim 2020: 54).

The constituent structure we propose (58) takes this into account combining PreGen and empty Det first, and the complex DetP with N later. NP approaches in HPSG (cf. Pollard & Sag 1994: 53; Ginzburg & Sag 2000: 193; Kim 2020: Section 3.1; for English), but also some DP analyses for German (Karnowski & Pafel 2004: 181–184; Hartmann & Zimmermann 2003: 180; a.o.) support this structure reflecting the distributional properties of the prenominal constituent and the empirical fact that PreGen and empty Det together determine the denotation of N. Therefore, our approach accounts for the data leading to a more parsimonious analysis. Determiners and complex prenominal structures are accounted for with the same syntactic machinery, the parallel between NPs and sentences still holds, and the constituent structure supported by data and literature is reflected in our analysis.

First, we have given an outline of how the DP analysis developed (Section 2). The history of the DP analysis helps us understand how changes in the theoretical axioms of a framework lead to changes in the analysis of NCs. This is particularly important as this development generates our current research question: DP or NP. In Section 3, we discussed some DP analyses for German that take into account the realisation of PreGens (Olsen 1991; Hartmann & Zimmermann 2003; Sternefeld 2009; 2015; a.o.). We focussed on (i) the type of element selected as determiner, (ii) how theta roles are assigned and which assumptions are needed to account for arguments inside NCs, and (iii) which problems a theory of local selection (Sag 2007; 2012) encounters in DP analyses. In Section 4, we have given a brief HPSG introduction in order to show the differences (w.r.t. MGG) that allow a local NP analysis. As our analysis shows, most of the tasks functional heads in MGG deal with are handled in the AVMs of lexical heads, enabling more surface-oriented approaches. In Section 5, we provided a detailed and formalised analysis of German NPs. First (Section 5.1), we concentrated on nominalisation of verbal stems and the (verb to noun) argument inheritance accounting for the mapping between argument structure and syntactic valency of lexemes. Second (Section 5.2), we presented the generalisation for case assignment in the verbal and nominal domain taking also the genitive vs. von-PP variation in German into account. Third (Section 5.3), concentrating on the prenominal position, we showed a similar structure in some German dialects: the prenominal dative with possessive. Moreover, we motivated the analysis of possessive elements as nominal arguments accounting for a uniform treatment of argumental NPs and possessives, hence giving an adequate description of the co-occurrence and constituent order restrictions. Fourth (Section 5.4 and 5.5), we showed how the syntactic and semantic combination is modelled. We provided the phrasal constraints needed to license the structures at hand only recurring to the standard phrasal constraints used in the framework. Furthermore, we developed a semantic account able to provide the interpretation of PreGens also in complex and recursive structures as they are found in corpora. In Section 6, we discussed an alternative NP analysis in HPSG: the functor approach, and discussed the possibility to account for NPs with a raising analysis based on weak heads. We compared the advantages and drawbacks of both analyses and concluded that our analysis could best explain the data. As a reviewer mentioned, the DP vs. NP debate does not necessarily concern headedness alone, but it comprises aspects of a functor-argument relation as well and this is exactly what our analysis reflects. On the one hand, the N head projects its properties determining its phrasal distribution. Furthermore, it determines which elements it can be combine with, which position these elements can have, and how they are interpreted. On the other hand, the determiner is acting as a functor on a semantic level (cf. Copestake et al. 2001: 145) taking PreGen and N head as arguments (cf. (67)). That is, our approach allows the double selection (Det selects N, N selects DetP)73 needed to account for the data presented without neglecting the head status of N, which is the central question of the DP-NP debate.