2.1 The correlate criterion

In Hansen & Heltoft's (2011) framework, relative clauses are referred to, interchangeably, as relative clauses and as ‘implicative’ clauses (e.g. Hansen & Heltoft Reference Hansen and Heltoft2011:1515ff.; see also Mortensen Reference Mortensen, Durst-Andersen and Müller2011b), a notion originating from Diderichsen (Reference Diderichsen1946) and employed also in e.g. Faarlund et al. (Reference Faarlund, Lie and Vannebo1997). Implicative (sub)clauses are defined as follows:

This constituent, shared by subclause and superordinate clause, is expressed by what is referred to as a correlate, placed in the superordinate clause, prototypically realised by a nominal (Hansen & Heltoft Reference Hansen and Heltoft2011:1516), as in (3).

1. (3)

   

The above subclause can thus be said to have a semantically empty subject slot which is ‘filled’ by the referent of the antecedent noun hund ‘dog’. This way of defining relative clauses, which I refer to as the correlate criterion, is widespread, not only in Scandinavian grammar. Thus, we find similar definitions and characterisations in e.g. Givón (Reference Givón1990:650ff.), Faarlund et al. (1997:1054), Van Valin & LaPolla (Reference Van Valin and LaPolla1997:590ff.), De Vries (Reference De Vries2002:14), Huddleston, Pullum & Peterson (Reference Huddleston, Pullum and Peterson2002:1036) and, as quoted in the following passage, Teleman et al. (Reference Teleman, Hellberg and Andersson1999b):

2.2 The subordinator criterion

In Teleman et al.'s (1999b) definition above, an additional criterion or characteristic is specified, namely the nature of the subordinator introducing the subclause. Most accounts of relative clauses involve such reference to more or less specific sets of subordinators said to introduce relative clauses, sometimes given as actual criteria, sometimes merely as prototypical characteristics of relative clauses, e.g.:

Associating relative clauses closely with ‘relative pronouns’ is a well-established tradition also in Danish grammar (e.g. Wiwel Reference Wiwel1901, Mikkelsen Reference Mikkelsen1911, Jespersen Reference Jespersen1939–40, Diderichsen Reference Diderichsen1946, Hansen Reference Hansen1967), even though in modern Danish, as well as in Swedish and Norwegian, relative pronouns are hardly a prototypical feature of relative clauses. As Hansen & Heltoft (Reference Hansen and Heltoft2011:1522) point out, subclauses modifying nouns or pronouns are primarily introduced by the conjunction som/ØFootnote ³ (‘which’/‘that’/‘who(m)’/Ø), rather than by hv-pronouns (‘wh’-pronouns), which mostly occur in archaic language or extremely formal style, as in (4).

1. (4)

   

As I will discuss later, hv-pronouns certainly do occur in Scandinavian relative clauses, in Danish especially hvor ‘where’ in locative and temporal relative clauses, as well as a few hv-pronouns with predicational or sentential correlates, as in (5).

1. (5)

   

However, there is no synchronic evidence to suggest that Scandinavian relative clauses are prototypically introduced by relative pronouns, which is probably what leads many Scandinavian grammars to incorporate both the correlate criterion and the subordinator criterion when accounting for relative clauses.

2.3 Other approaches

In other accounts, relative clauses are defined or characterised without the requirement of a correlate relation or certain subordinators, e.g.:

In cross-linguistic and typological frameworks, relative clauses are often defined in terms of broader notional or semantic characteristics (e.g. Dik Reference Dik1997b, Comrie Reference Comrie1998, Givón Reference Givón2001, Cristofaro Reference Cristofaro2003, Andrews Reference Andrews and Shopen2007), e.g.:

These other approaches to relative clauses seem to approximate the general functional understanding of attributive clauses, as initially outlined, insofar as they base the relative clause category roughly on the criterion that the subclause, in its realised construction, somehow modifies a superordinate argument constituent, which is more or less equivalent to saying that it functions as an attribute. These terminological overlaps between related but nonetheless different empirical phenomena contribute to a general blurring of what constitutes a relative clause. However, in the following I will concentrate on the subordinator criterion and the correlate criterion, as they are the traditional ones typically featured in, among others, Scandinavian grammars, and the ones I find truly problematic.

3.1 Demarcating ‘relative subordinators’

As outlined above, the prototypical relative clause in Danish, Swedish and Norwegian is the som/Ø-clause, and apart from the conjunction som, a number of hv-pronouns (v-pronouns in Swedish) are usually part of the formal criteria, explicitly or implicitly, for identifying relative clauses. Because som/Ø in the Scandinavian languages has taken over most of the work done by relative pronouns in other languages, or earlier Scandinavian stages, this conjunction is commonly considered one of the relative subordinators. However, while in Danish it can be distinguished by its general omissibility in relative constructions (see e.g. Mortensen Reference Mortensen2011a), som alone is not a decisive signal of ‘relativeness’, as subclauses introduced by som may alternatively function as adverbial clauses with e.g. temporal or comparative meanings. Similarly, so-called relative pronouns are not inherently relative, as they can also introduce interrogative clauses (in which case they are often referred to as interrogative pronouns). Tradition and diachronic arguments aside, what determines, then, which subordinators can be considered relative subordinators? Huddleston & Pullum characterise relative clauses as containing ‘within their structure an anaphoric element whose interpretation is determined by the antecedent’ (Huddleston et al. Reference Huddleston and Pullum2002:1034), although, as they acknowledge, the question whether this element is in fact anaphoric (in the case of relative clauses) or has a different referential status (in the case of interrogative clauses) is determined by other factors of the construction, as well as by pragmatics. Moreover, many linguistic forms other than those considered relative subordinators can be ‘anaphoric elements whose interpretation is determined by the antecedent’. For example, Huddleston & Pullum (Reference Huddleston and Pullum2002) include when and while, which in Danish happen to translate into the conjunctions når/da and mens (see (6a); example (6b) will be referred to in the next paragraph), which are hardly recognisable as pronouns, yet are commonly considered to have the same potential for introducing ‘adverbial relative clauses’ (e.g. Diderichsen Reference Diderichsen1957[1946]:210ff.; Hansen & Heltoft Reference Hansen and Heltoft2011:1530ff.), to which I will return.

1. (6)

   

Whereas in English a fairly homogenous group of ‘relative subordinators’ can be formed by wh-pronouns (with that as the only exception), defining a clearly demarcated and/or homogenous set of relative subordinators in Danish, as well as in Swedish and Norwegian, is not easy, given the fact that the hv-/v-form cannot be an absolute criterion. Indeed, Hansen & Heltoft include time clauses in general as a kind of relative clauses (Hansen & Heltoft Reference Hansen and Heltoft2011:1531), which is a further widening of the category. While I do not fully agree with their analysis, as I will discuss later, it is true, also for English, for example, that time clauses, as well as many other kinds of adverbial-like clauses, can function exactly like relative clauses, e.g. in (6b) above. Since the subordinator criterion alone is thus difficult to apply in a consistent and empirically justifiable manner, at least in the Scandinavian context, grammars tend to combine it with the correlate criterion in order to account for the properties of relative clauses. I will discuss the correlate criterion in what follows.

3.2 What is (not) a correlate relation?

As illustrated in example (3) above, when dealing with prototypical relative clauses such as som/Ø-clauses, the empirical data seem to match both the subordinator criterion and the correlate criterion quite well. However, relative clauses with less prototypical correlates and/or subordinators are also part of the picture, leading to a radial-like organisation of the category. For instance, less prototypical cases are presented in many grammars, including constructions with adverbial or sentential correlates, and even correlates with no manifest representation (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985, Faarlund et al. Reference Faarlund, Lie and Vannebo1997, Teleman et al. Reference Teleman, Hellberg and Andersson1999b, Hansen & Heltoft Reference Hansen and Heltoft2011), as in (7), (8) and (9).

1. (7)

   

2. (8)

   

3. (9)

   

In fact, Faarlund et al. (Reference Faarlund, Lie and Vannebo1997) and Hansen & Heltoft (Reference Hansen and Heltoft2011) even structure their presentations of relative clauses according to the status of the correlate in the superordinate clause, from prototypical to less prototypical cases, as the following illustrates:

Correlate represented obligatorily

• • Nominal correlate (i.e. prototypical adnominal relative clauses)

• • Predicate correlate (i.e. predicational or sentential relative clauses)

Correlate represented optionally

• • Adverbial correlate (i.e. adverbial relative clauses)

Correlate not represented

• • ‘General’ correlate (i.e. nominal/headless relative clauses, see Section 3.3 below)

  (Hansen & Heltoft Reference Hansen and Heltoft2011:1516, my translation)

In cases of non-represented correlates, the correlate is held to be implicitly present in the superordinate clause, meaning for instance that the subclauses in (10) and (11) below are both classified as (adverbial) relative/implicative clauses even though only (10) features an explicit correlate.

1. (10)

   

2. (11)

   

Similarly the subclauses in (12) and (13) below are both classified as relative clauses even though only (12) features a correlate.

1. (12)

   

2. (13)

   

As noted, a basic criterion for relative clauses is the representation of a correlate in the superordinate clause, which questions the inclusion of (11) and (13) as relative clauses. Seeing them as relative clauses must be motivated by the other traditional criterion, i.e. the subordinator criterion, given that they are introduced by the ‘relative subordinators’ da and vad, respectively. As discussed, however, many kinds of subordinators seem to ultimately qualify as relative subordinators, especially in the Scandinavian languages, which makes the status of the correlate criterion all the more important. In what follows I discuss the two less prototypical subgroups of relative clauses, i.e. nominal/headless relative clauses and adverbial relative clauses.

3.3 The notion of nominal relative clauses

Nominal (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985) or headless (e.g. Givón Reference Givón2001) relative clauses, in the Danish tradition referred to as ‘almene relativsætninger’ (‘general relative clauses’; Hansen & Heltoft Reference Hansen and Heltoft2011:1538), differ, as their names indicate, from prototypical relative clauses in (at least) two respects: firstly, their correlate is obligatorily non-represented in the superordinate clause, and secondly, they do not function as attributes but as nominals, as illustrated in (14)–(16).

1. (14) You’re not Ø [who I thought you were]. (compare ‘. . . the person I thought you were’)

   (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1057)

2. (15)

   

3. (16)

   

Fundamentally, the very acceptance of non-represented correlates more or less undermines the empirical basis of the relative clause category, as long as relative clauses are defined by featuring a correlate in the superordinate clause. The fact that nominal relative clauses, as they are called here, are nevertheless treated as actual relative clauses (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1056; Faarlund et al. Reference Faarlund, Lie and Vannebo1997:1058; Teleman et al. Reference Teleman, Hellberg and Andersson1999b:506; Hansen & Heltoft Reference Hansen and Heltoft2011:1486) probably derives from their being introduced by hv-/v-/wh-pronouns, as discussed above. Still, the reason why subclauses introduced by such pronouns are called relative clauses in the first place is exactly that the pronoun relates to a superordinate antecedent and is determined by this antecedent (e.g. Huddleston et al. Reference Huddleston and Pullum2002:1034). Thus, when the antecedent is not present, the status of the pronoun is different too, as it can no longer be said to have anaphorical reference. In this respect, as discussed in e.g. Jespersen (Reference Jespersen1927:72ff.), Quirk et al. (1985:1056ff.) and Dik (Reference Dik1997b:80ff.), nominal relative clauses resemble interrogative subclauses, with which they share not only the initial pronouns, but also the nominal function. In order to distinguish between them, various relative vs. interrogative characteristics are often referred to in the literature, for instance in the form of vague semantic characteristics as here:

Distributional differences are also occasionally mentioned, such as the fact that interrogative clauses can enter into cleft constructions, whereas nominal relative clauses cannot (e.g. Togeby Reference Togeby2003:97), as illustrated in (17a) and (17b).

1. (17)

   

However, these semantic and distributional differences are hardly clear-cut reasons for the traditional separation of the subclauses into two entirely different groups. An alternative approach is to treat them, like e.g. Wiwel (Reference Wiwel1901:323), collectively as simply ‘pronominal clauses’Footnote ⁵ that may pragmatically be used in different ways. In fact, the semantic difference hinted at by Quirk et al. above can be paraphrased as one of different degrees of specificity, i.e. a more general, pragmatic and not structural, aspect of referring (see e.g. Preisler Reference Preisler1997:163ff.). To give a general example of the specificity domain, the indefinite nominal phrase a girl can be used for referring more or less specifically to its referents, e.g. in (18)–(20).

1. (18) Diamonds are a girl's best friend.   (generic reference: ‘girls as such’)

2. (19) I’d love to meet a girl and settle down.   (unspecific reference: ‘some girl’)

3. (20) I met a girl yesterday.   (specific reference: ‘a particular girl’)

Correspondingly, whether a pronominal subclause is used as an interrogative clause or as a nominal relative clause is essentially a question of its referential status, that is, the degree of specificity of the pronoun, as illustrated in (21)–(23).

1. (21) Please tell me [what you do].   (unspecific reference → interrogative)

2. (22) I appreciate [what you do].   (specific reference → relative)

3. (23) I’m interested in [what you do].   (unspecific or specific reference → interrogative or relative)

Thus, in (21), the subclause content – ‘what you do’ – does not refer to a specified job or activity, i.e. it has unspecific reference, whereas the one in (22) refers to a specific job or activity already known by the speaker. (23) is ambiguous and may be taken either as a question or as a statement. The content and expression structure of the subclauses are the same; it is only the contextually determined specificity of reference which arguably differs. These differences of pragmatic usage indeed also explain the distributional differences observed by e.g. Togeby (Reference Togeby2003). While the differences seem to lead many traditional grammars into dividing such subclauses into separate groups of nominal relative clauses and interrogative clauses, this division is hardly warranted, as much as we do not need to divide the indefinite article a into different classes because of its different uses exemplified in (18)–(20).

In other words, nominal relative clauses have at least as much in common with interrogative clauses as with other relative clauses, and might as well be dealt with as part of a shared category such as Wiwel's (1901) ‘pronominal clauses’.Footnote ⁶ Crucially, however, if one insists on treating nominal relative clauses as a kind of relative clauses, they cannot at the same time be regarded as nominal clauses, since the assumed implicit correlate in the superordinate clause, which would warrant the relative clause label altogether, has the syntactic function of a phrase head (see e.g. Faarlund et al. Reference Faarlund, Lie and Vannebo1997:1048; Becker-Christensen Reference Becker-Christensen2010:142; Hansen & Heltoft Reference Hansen and Heltoft2011:1515) with the subclause as its postmodifier, rendering the subclause an attributive clause, not a nominal one. Still, since these clauses are in fact nominal and not attributive, as long as no implicit constituents are forced into the analysis, their inclusion as relative clauses is hardly justified, from a Scandinavian as well as a general perspective.

3.4 The notion of adverbial relative clauses

The possibility of adverbial or adverbial-like relative clauses, as illustrated in (24) and (25), is accounted for in several grammars, e.g. (Diderichsen Reference Diderichsen1957[1946]:210ff.; Faarlund et al. Reference Faarlund, Lie and Vannebo1997:1064ff.; Teleman et al. Reference Teleman, Hellberg and Andersson1999b:507ff.; Huddleston et al. Reference Huddleston and Pullum2002:1051; Hansen & Heltoft Reference Hansen and Heltoft2011:1530ff.).

1. (24)

   

2. (25)

   

However, as also mentioned in connection with example (11) above, such time clauses and certain other adverbial-like subclauses are commonly treated as relative clauses even when there is no correlate represented in the superordinate clause (e.g. Faarlund et al. Reference Faarlund, Lie and Vannebo1997:1061; Teleman et al. Reference Teleman, Hellberg and Andersson1999b:508; Hansen & Heltoft Reference Hansen and Heltoft2011:1531). Analysing such subclauses as inherently relative parallels the treatment of nominal relative clauses, in that the implicit representation of a correlate in the superordinate clause is the classification basis. Still, this analysis is not empirically grounded and it arguably over-complicates the description of what might otherwise simply be treated as adverbial clauses, as would be the appropriate classification of, for example, the subclause in (26).

1. (26)

   

If an underlying motivation for such analyses is that the subordinating conjunction, in this case da ‘when’, is seen as a correlate constituent in the subclause (as e.g. suggested for Swedish equivalents in Teleman et al. Reference Teleman, Hellberg and Andersson1999b:508), this arguably opens up for the possibility of any kind of subordinating conjunction to be treated as a correlate. Furthermore, the conjunction can hardly be said to co-refer with a constituent in the superordinate clause, as required by the relative clause definitions – rather, the entire subclause event can be said to coincide with the entire superordinate event, which only confirms that we are dealing with an adverbial relation of simultaneity. There is no obvious reason why the adverbial relation should also be treated as a relative relation.

Further, in cases when adverbial-like clauses do correlate with a manifest head constituent in the superordinate clause, they arguably resemble appositional (also known as appositive) clauses more than relative clauses. These two main groups of attributive clauses are traditionally distinguished according to differences in the subclauses’ internal make-up, in that an empty slot is said to characterise prototypical relative clauses (as part of the correlate criterion), whereas appositional clauses do not feature such empty slot, as illustrated in (27).

1. (27)

   

Applying this criterion, it is hard to see how so-called adverbial relative clauses can be said to have an empty slot and thus qualify as relative clauses. Rather, they are complete subclauses that specify or otherwise modify a superordinate element, in constructions parallel to those of traditional appositional clauses, which may be illustrated by comparing (27) above with any ‘adverbial relative clause’ example, e.g. Hansen & Heltoft's in (28).

1. (28)

   

In any case, an important final point must be made which echoes what was said concerning nominal relative clauses: if adverbial subclauses like the one in (26) above without an explicit correlate relation are treated as relative clauses, they cannot at the same time be regarded as adverbial clauses, since any superordinate correlate, even if implicit, has the syntactic status of a phrase head, rendering the subclause an attributive clause. Thus, when analysing a construction like (26), one must ultimately choose between treating the subclause as an adverbial clause or as a relative and thereby attributive clause, in which case the simple adverbial analysis seems the appropriate one.

In other words, the kind of semantic link connecting a relative clause to its correlate, whether implicit or manifest, is essentially the same kind of link connecting an attributive clause to its head. This link is brought about by means of syntactic construction and is not an inherent property of certain subclauses or certain subordinators. Thus, at least for the Scandinavian languages, relative clauses can hardly be regarded as a formal subclause category, but are rather a functionally defined class like that of attributive clauses. This can be illustrated by first looking at the traditional relative clause forms in Danish and the main formal and functional features traditionally said to characterise them, presented in Table 1.

Table 1.

Traditional view of Danish relative clauses and their main features.

Table 1 reflects the traditional understanding of relative clauses as a versatile yet coherent group of subclauses with prototypical and less prototypical members, all identified and characterised by formal features such as particular subordinators and various kinds of correlate relations realising different syntactic functions.

However, if the formal features are examined and used consistently as relative clause criteria, it is clear that the so-called relative subordinators are not ‘relative’ per se, as they can all introduce non-relative clauses as well, except perhaps for som/Ø, which in Danish may be structurally distinguished by its omissibility. Further, if the correlate criterion is to make sense, only constructions featuring some kind of manifest correlate can be labelled relative constructions – otherwise anything could be said to implicitly satisfy the criterion. Consequently, since the nature of a correlate is essentially that of a phrase head, correlate constructions are necessarily attributive constructions, which leads to a revised view of relative clause features, as presented in Table 2.

Table 2.

Revised view of Danish relative clauses and their main features.

In this more consistent view of relative clauses, the only clause type that fully satisfies the traditional criteria is arguably som/Ø-clauses, as discussed, and the only features shared by all the clause types are the basic criterion that they are part of explicit correlate constructions and, as a direct implication of this, that they function as attributes (in some cases arguably as attributes to full predicates or sentences). In other words, what essentially characterises the relative clauses is that they are attributive clauses.

3.5 Is the category of relative clauses justified for modern Scandinavian languages?

Provided that the only prototypical member of the relative clause category in the Scandinavian languages seems to be som/Ø-clauses, the category as traditionally outlined is densely overpopulated by members that do not fully, if at all, meet its basic criteria. Moreover, since ‘relativeness’ can be boiled down to merely a way of constructing subclauses that amounts to syntactic attribution, the classificatory work offered by the relative clause category is already taken care of, as it were, by the traditional attributive clause category. The only difference is that the correlate criterion and the subordinator criterion formally excludes appositional clauses – otherwise, relative clauses and attributive clauses are essentially the same category. At the same time, as illustrated, consistent and exhaustive criteria for a coherent group of relative clauses are not, and could hardly be, formulated to adequately match modern Danish, Swedish and Norwegian.

A more fruitful and consistent way to go about a formal sub-categorisation of this empirical domain might be to approach it top down, that is, to explore first the syntactic relation constituting the attributive function, and then which (groups of) subclauses, identified formally e.g. by their subordinators, can actually occur in this function and under which circumstances. One example of such subgroup is what I have referred to as pronominal clauses, which more consistently than the relative clause category cover e.g. Danish hv-subclauses. In the following, I propose a general description of attributive clauses in Danish, starting from an overall discussion and account of the attributive clause category.

4.1 Phrase constituents, not clause constituents

As discussed initially, most accounts of attributive clauses focus on nominal phrases in which a subclause functions as postmodifier. However, specific criteria marking the boundaries of the attributive clause category against other syntactic constructions are rarely given; yet one can generally gather that clauses treated as attributive clauses are (postmodifying) parts of other constituents, forming complex phrases with them, rather than sentence constituents in their own right. If we consider this a criterion, there is no reason why only nominal phrases should be able to feature attributive clauses. Indeed, examples have already been given of adverbial phrases containing attributive clauses, and examples involving adjectival phrases will be discussed later. Thus, I regard subclauses that are not themselves full constituents, but postmodifiers in phrases of any kind, as attributive clauses. A special kind of attributive clauses, then, are those that do not postmodify heads of phrases but full predicates or sentences. For this reason, among others, I consider attributive clauses a more appropriate term than adjectival clauses, since ‘adjectival’ is traditionally associated specifically with noun modification.

To determine that a subclause is not a constituent in its own right, a topological guideline can be employed: Since a sentence constituent can normally only be fronted as a whole, attributive clauses, as opposed to nominal and adverbial clauses, cannot be fronted alone – they need their governing head, as in (29a–c).

1. (29)

   

4.2 Non-obligatory postmodifiers

The head of a phrase can be premodified and postmodified by words, phrases and, in the case of postmodifiers, full clauses, constituting attributive clauses. An example of a nominal phrase featuring both premodifiers and an attributive clause was given in (1) and is repeated in (30) below.

1. (30)

   

Postmodification of a head (or full predicate or sentence) is always structurally non-obligatory, while the postmodifying elements may be semantically/pragmatically bound to various degrees. Bound postmodifiers are in many cases referred to as complements (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:63; Teleman, Hellberg & Andersson Reference Teleman, Hellberg and Andersson1999a:159; Pullum & Huddleston Reference Pullum and Huddleston2002:24), although no adequate structural distinction can be made between such complements and unbound postmodifiers, and the practical application of the distinction is not clear. For example, Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985) describe the postmodifying subclause in (31) below as a complement to the head more, forming a comparative clause.

1. (31) John is more intelligent [than Bill is].   (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1145)

Huddleston & Pullum (Reference Huddleston and Pullum2002) also treat such comparative clauses as complements. Certainly, textual and informational circumstances may render an expectation of the addition than Bill is, but this is merely a question of what context to assume for the example. If an alternative preceding context is specified as in (32) below, the subclause can no longer be considered bound and may be omitted.

1. (32) Bill is not particularly bright. John is more intelligent Ø.

This means that syntactically, the subclause can hardly be classified as anything more specific than a postmodifying, hence attributive clause. Alternatively, if any element that is felt to be semantically/pragmatically bound is to be classified as a complement, we would also expect to see e.g. restrictive relative clauses treated as complement clauses; yet both Quirk et al. and Huddleston & Pullum treat restrictive relative clauses simply as postmodifiers (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1239; Huddleston et al. Reference Huddleston and Pullum2002:1034), in line with traditional grammar.

Thus, since there is hardly a clear syntactic distinction to be made between attributive clauses and complement clauses modifying phrase heads (or full predicational/sentential heads), I shall regard any syntactically non-obligatory postmodifying subclause as an attributive clause, including certain postmodifying clause types referred to as complement clauses in other frameworks.

What follows is an outline of the types of constructions that are able to feature attributive clauses in Danish, namely nominal phrases, adverbial phrases and adjectival phrases, as well as predicational and sentential structures as a peripheral kind.

5.1 Nominal phrases

The most well-established group of phrases containing attributive clauses are those headed by nouns and pronouns of various kinds. They hardly require in-depth introduction as they have already been discussed and illustrated throughout the article, but a couple of examples follow.

1. (33)

   

2. (34)

   

5.2 Adverbial phrases

Adverbial phrases headed by adverbs or transcategorised nouns or adjectives may also contain attributive clauses. Such constructions include e.g. comparative and degree marking clauses, as in (35) and (36).

1. (35)

   

2. (36)

   

In constructions like (36), the degree adverb så ‘so’ may be considered the head of the attributive clause. Together they form a degree adverbial phrase modifying smukt ‘beautifully’, which is the head of the overall manner adverbial phrase.

Time, manner and place adverbs/adverbial constructions may also function as heads of adverbial phrases featuring attributive clauses, as in (37).

1. (37)

   

5.3 Adjectival phrases

Moreover, adjectival phrases, often headed by adjectives in the comparative, can also be modified by certain kinds of attributive clauses, as in (38).

1. (38)

   

Comparative clauses like the one above are occasionally treated as a separate class of subclauses in opposition to the function-defined classes (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1047ff.), but the ‘comparative’ feature is a purely semantic one. What makes such clauses comparative is the fact that they modify adjectives, for example, and by virtue of modifying a superordinate constituent, thus realising an (adjectival) phrase, they qualify as attributive clauses.

It is worth adding that in e.g. Swedish, Norwegian and English – though not in Danish – ‘emotive’ adjectives are able to function as heads for attributive clauses in constructions like the ones in (39) and (40).

1. (39)

   

2. (40)

   

In some frameworks, such clauses are treated as complement/bound clauses (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1049; Teleman et al. Reference Teleman, Hellberg and Andersson1999b:533; Huddleston Reference Huddleston2002:964), but Preisler (Reference Preisler1997) classifies them as postmodifiers as opposed to e.g. complements (Preisler Reference Preisler1997:51). As they are syntactically non-obligatory, I follow Preisler and treat them as attributive clauses. However, this kind of attributive construction is not observed in modern Danish, where a preposition is generally required after the adjective,Footnote ⁷ meaning that such subclauses are nominal rather than attributive, as shown in (41a) and (41b).

1. (41)

   

5.4 Predicational and sentential ‘heads’

Finally, parts of sentences as well as full sentences are capable of being modified in an attributive-like relation by certain kinds of subclauses (see e.g. Theilgaard Reference Theilgaard2009, Hansen & Heltoft Reference Hansen and Heltoft2011), as in (42)–(44).

1. (42)

   

2. (43)

   

3. (44)

   

I return to these predicational/sentential constructions later, and propose that they may be regarded as a peripheral type of head material.

I have given examples illustrating that nouns, pronouns, adjectives and adverbs as well as larger linguistic units can all function as heads for attributive clauses, forming nominal, adverbial and adjectival phrases, or full clauses. Still, different kinds of heads put different semantic restrictions on the clauses they govern, and vice versa, just as certain clause types are generally unable to function as attributive clauses. In the following, I examine a range of Danish subclause types, identified first and foremost by the subordinators introducing them, and account for the types of heads they are able to modify.

6.1 Som/Ø-clauses

As discussed, the central attributive clause in Danish is arguably the som/Ø-clause (‘which’/‘that’/‘who(m)’/Ø-clause). The fact that the conjunction som is generally optional in DanishFootnote ⁸ sets som/Ø-clauses apart from som-clauses, i.e. clauses of comparison or time. As opposed to som-clauses, som/Ø-clauses only occur as attributive clauses, modifying nouns and pronouns of all kinds, as in (45) and (46).

1. (45)

   

2. (46)

   

6.2 Hv-clauses

As discussed, Danish hv-clauses (‘wh’-clauses), which may collectively be labelled pronominal clauses, can be used for several syntactic and semantic purposes. ‘Interrogative’ hv-clauses are perhaps canonically nominal, but they do occur as attributive clauses as well in appositional-like constructions, governed by a few kinds of nouns/nominalisations expressing doubt, as in (47) and (48).

1. (47)

   

2. (48)

   

Further, in special contexts where highly formal or archaic language is used, a few hv-clauses, notably hvem- and hvad-clauses (‘who(m)’-clauses and ‘what’/‘which’-clauses), and clauses introduced by the possessive pronoun hvis ‘whose’, can function much like som/Ø-clauses, modifying nouns and pronouns, as in (49).

1. (49)

   

Moreover, hv-subordinators combined with prepositions can modify nouns and pronouns, also mainly in formal or archaic-style contexts, as illustrated in (50).

1. (50)

   

These are all constructions in which hv-clauses modify nouns or pronouns. As discussed, some hv-clauses can realise ‘sentential attributive clauses’, modifying the entire superordinate clause or a predicate contained in it. In Section 5.4 above, I gave examples of hvilket-clauses (‘which’-clauses); other significant types include e.g. hvorfor-clauses (‘why’-clauses) and hvorimod-clauses (‘whereas’-clauses), as in (51a) and (52a).

1. (51)

   

2. (52)

   

Some accounts treat such subclauses as adverbial or adverbial-like clauses (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985, Togeby Reference Togeby2003, Hansen & Heltoft Reference Hansen and Heltoft2011; but see discussion in Theilgaard Reference Theilgaard2009:55–56). However, this classification is unsatisfactory as the clauses lack the ability to be fronted, as shown in (51b) and (52b) above, an ability which characterises other adverbial clauses in Danish (e.g. Hansen & Heltoft Reference Hansen and Heltoft2011:1541).

Since these clauses do not modify single heads but entire clauses, it is impossible to determine whether they could be fronted along with their ‘head’, as in the case of other attributive clauses. Still, the fact that they cannot be fronted separatelyFootnote ⁹ might be seen as evidence in favour of treating them as attributive clauses, since attributive clauses are always postmodifiers as opposed to e.g. adverbial clauses. While Theilgaard argues that they do not have a constituent function at all (Theilgaard Reference Theilgaard2009:61), I find it reasonable to treat them as peripheral members of the attributive clause category.

The distribution of the attributive hv-clauses differ, such that some can only express interrogative meaning, modifying a few specific noun heads, e.g. hvordan-clauses (‘how’-clauses). Others can modify most kinds of nouns and pronouns as well as (parts of) full sentences, e.g. hvad-clauses (‘what’/‘which’-clauses), while some hv-clauses can only modify predicates and full sentences, e.g. hvorimod-clauses (‘whereas’-clauses). A final hv-clause is especially versatile, namely the hvor-clause (roughly: ‘where’/‘when’-clause), which is able to occur not only in all of the functions mentioned above, but also as attribute to adverb heads, as in (53) and (54).

1. (53)

   

2. (54)

   

6.3 At/Ø- and om-clauses

The primary attributive function of at/Ø-clauses (‘that’/Ø-clauses) and om-clauses (‘if’/‘whether’-clauses) is that of apposition, as in (55) and (56), quite similar to interrogative attributive hv-clauses.

1. (55)

   

2. (56)

   

While om-clauses roughly share their distributional pattern with interrogative hv-clauses, at/Ø-clauses can modify a broader range of nouns referring to opinions, beliefs, statements etc. Similarly to som/Ø-clauses, at/Ø indicates that at is optional in various contexts, including attributive clauses, as illustrated in (57).

1. (57)

   

Finally, as discussed in Section 5.2 above, at/Ø-clauses can modify adverb heads in the characteristic degree marking construction. Note that also så-clauses (‘so (that)’-clauses) can occur in this kind of construction, e.g. in (58).

1. (58)

   

6.4 End-clauses

End-clauses (‘than’-clauses) represent one of the few clause types apart from som/Ø-clauses that can only function as attributes (see e.g. Becker-Christensen Reference Becker-Christensen2010:144). They do so as modifiers of e.g. comparative adverbs, adjectives and other elements involved in comparisons or distinguishing relations, and thus may be part of adjectival, adverbial or nominal phrases, the former and the latter exemplified in (59) and (60).

1. (59)

   

2. (60)

   

6.5 Time, place and manner clauses, etc.

A range of subclauses that are typically recognised as adverbial clauses can also occur in the attributive function. For instance, attributive time clauses were discussed earlier – and while I oppose to the tradition of treating time clauses as inherently ‘relative’, it is indeed relevant to point out their ability to function in this way, i.e. as attributive clauses. Time clauses are not the only ‘adverbial-like’ subclauses that can occur as attributes. In fact, any subclause that can function as an adjunct clause, as opposed to a disjunct clause (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985:1070ff.), can also have attributive function (see Mortensen Reference Mortensen2011a for a detailed distributional analysis of such subclauses). For instance, a typical time clause such as a mens-clause (‘while’-clause) can occur as adjunct and as attribute, as illustrated in (61) and (62).

1. (61)

   

2. (62)

   

Subclauses like these designate concrete, objective circumstances of what is expressed in the superordinate clause, such as time, place, manner, cause, condition etc. Disjunct clause forms, on the other hand, as in (63a), expressing speaker-subjective circumstances such as reason, concession, justification, contrast etc., cannot be constructed with attributive function, as illustrated in (63b).

1. (63)

   

Even when the selvom-clause (‘although’-clause) is forced into a position adjacent to a potential head constituent as in (63b), the syntactic relation between them is not one of attribution. Instead, the selvom-clause must be interpreted as a parenthetical addition modifying the entire superordinate clause in an adverbial relation, which is illustrated by the fact that the subclause can easily be fronted separately, as in (63c) above.

A number of subordinators can introduce both adjunct and disjunct clauses, including a range of conjunctions featuring a temporal as well as a non-temporal variant. Such variants have different semantic and syntactic properties reflecting how closely they are integrated in the superordinate nucleus (see e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985; Haegeman Reference Haegeman, Zanuttini, Campos, Herburger and Portner2006; Mortensen Reference Mortensen2006, Reference Mortensen2011a; Verstraete Reference Verstraete2007; Hansen & Heltoft Reference Hansen and Heltoft2011), including the distributional difference that only potential adjunct clauses are able to function as attributes.

Moreover, due to specific lexical differences, certain adjunct-like clause types are more easily constructed as attributive clauses than others. For instance, most time clauses readily modify nouns and time adverbials of all kinds, as discussed and illustrated in Section 3.4 above, whereas e.g. causal fordi-clauses (‘because’-clauses) require more specific types of heads, as illustrated in (64) to (67).

1. (64)

   

2. (65)

   

3. (66)

   

4. (67)

   

Thus, while the time clauses above all modify heads designating more or less concrete entities or specifications, the complexity of a causal relation like the one in (67) requires that a predication or a similarly complex entity be incorporated within the head constituent. For this reason, attributive causal clauses and other semantically complex clause types seem to be restricted to nominal phrases headed by certain kinds of nouns, seeing that e.g. adverb and adjective heads can hardly express predicational content. Glæde ‘joy’, in the above case, can for instance be said to incorporate a predication such as at være glad ‘being joyful’, to which the modifying subclauses express different causes. Note that analysing the fordi-clauses as adverbial clauses would be inappropriate: even if glæde is seen as part of the preceding clause, the causal relations cannot be interclausal – only glæde(n) is modified.

Similar kinds of complex predicational content can also be expressed through for instance nominalisations, as in (68).

1. (68)

   

Similar preferences apply to other semantically complex adjunct-like clauses such as e.g. hvis-clauses (‘if’-clauses) and så-clauses (‘so (that)’-clauses), as in (69) and (70).

1. (69)

   

2. (70)

   

Finally, the general fact that potential adjunct clauses can modify predicational content both in terms of adverbial modification and in terms of attributive modification calls for a clarification of the boundary between the two kinds of syntactic relations. For instance, it is not necessarily clear-cut whether the mens-clause (‘while’-clause) in (71) is an adverbial clause or an attributive clause.

1. (71)

   

Yet, considering that et simpelt tyveri af en smartphone ‘a simple theft of a smartphone’ has the structure of a nominal phrase (i.e. the prepositional object of ved ‘in the case of’), the subclause is most appropriately classified as an attributive clause modifying the head tyveri ‘theft’. Still, this construction seems closer to an adverbial construction than most of the previous attributive constructions discussed, due to the fact that the head is part of a full predication, i.e. a semantic unit normally modified by e.g. time adverbials. The attributive relation adds to a densely packaged information unit suited for specifying a certain ‘type of damage’. Other kinds of cases that might be considered borderline examples, include superordinate infinitives modified by adjunctive clauses, as in (72).

1. (72)

   

The mens-clause resembles an attributive clause in terms of functioning as a modifier within a phrase, i.e. an infinitive phrase. However, the criterion that an attributive clause must be non-obligatory is hardly met in this case, as leaving out the mens-clause would render the sentence somewhat incomplete, as illustrated in (73).

1. (73)

   

Rather, the mens-clause might be treated as a predicate satellite (see e.g. Dik Reference Dik1997a:51), i.e. a kind of bound adverbial clause constituting an integral part of the verbal activity itself. Such borderline examples are part of any classification approach and illustrate the inevitable fuzziness of grammatical categories. The success of a classification approach, then, should depend on whether it offers fewer problematic borderline cases and greater explanatory power than its alternatives.

The distributional patterns observed for the various subclause forms analysed throughout the article can be summarised in Table 3, in which the Xs illustrate which of the subclause types can occur as attributes in the three different phrase types and as part of predicational/sentential constructions.

Table 3.

Possible attributive clause constructions in Danish. (The fourth column is shaded to illustrate that predicational/sentential attributive constructions represent a peripheral case.)

Table 3 confirms that som/Ø is arguably the most general attributive subordinator in Danish, and that a nominal head is the prototypical base for attributive clauses. The fourth column is shaded to illustrate that predicational/sentential attributive constructions represent a peripheral case. Finally, the bottom row is void of Xs to illustrate that certain subclause forms/variants cannot occur as attributive clauses at all. This organisation of the distributional patterns highlights two prototypicity hierarchies, showing on the one hand e.g. that adjectival phrases (let alone predicational/sentential constructions) are the ones least capable of featuring attributive clauses, and on the other hand that e.g. hvis-clauses and fordi-clauses are the least versatile forms to occur as attributive clauses.