2. DANISH NON-SPECIFIC RELATIVES

Free relative constructions are different from ordinary relative clauses. A free relative construction contains a relative phrase which is ‘fused’ with the antecedent. If the free relative pronoun has a determiner function, the antecedent is explicitly present. This is illustrated in (1) where the antecedent and relative pronoun in parentheses on the left indicate the ‘unfused’ representation of the free relative pronoun in parentheses on the right.Footnote ¹

1. (1)

   

It should be noted that if the ‘fused’ relative pronoun functions as a subject, as in (1a) and (1b), an expletive obligatorily appears in subject position, see e.g. Bjerre (Reference Bjerre2013) for a discussion of the occurrence of the expletive subject in Danish interrogative and relative clauses.

Because of this dual function of the free relative pronoun, free relatives in many languages obey the so-called ‘matching’ effect, see Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978). This is also the case with Danish free relatives. A consequence of the ‘matching’ effect is that free relatives, in contrast to ordinary relative clauses, do not allow PP pied piping. This is illustrated for Danish in (2) and (3), where we can see that the argument from Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978) also holds for the equivalent examples in Danish.

1. (2)

   

2. (3)

   

In (3b) the verb genlæse ‘reread’ in the main clause requires an NP object. However, the free relative phrase is a PP and consequently there is a mismatch of category.

Semantically, the non-specific free relatives are commonly taken to involve universal quantification, e.g. Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978), Larson (Reference Larson1987), Tredinnick (Reference Tredinnick, Eckardt and van Geenhoven1994) and Iatridou & Varlokosta (Reference Iatridou, Varlokosta and Kusumoto1996), whereas specific free relatives are definite. We will not go into any detail regarding their semantics, but at least it can be noted that the non-specific free relatives usually admit a paraphrase involving a universal quantifier, as also shown in (1) above.Footnote ²

In this section, we will show a collection of Danish data containing non-specific free relatives. The Danish counterparts of the English free relative pronouns suffixed by -ever are multi-word free relative pronouns ending in som helst. The set of non-specific free relative pronouns includes hvem som helst ‘whoever’, hvad som helst ‘whatever’, hvor som helst ‘wherever’, når som helst ‘whenever’, and hvilken som helst ‘whatever/whichever’. In (4) we show non-specific free relatives in subject position. (The URLs for the sources of the examples in this sections are given in endnotes.)

1. (4)

   

Footnote ³Footnote ⁴Footnote ⁵ In (5) the examples contain non-specific free relatives in object position.

1. (5)

   

Footnote ⁶Footnote ⁷Footnote ⁸ And finally, in (6) the examples contain non-specific free relatives as adverbials.

1. (6)

   

Footnote ⁹Footnote ¹⁰Footnote ¹¹ Danish non-specific free relative pronouns may occur with bound relative clause sisters.Footnote ¹²

In (7) the free relative phrases are followed by bound relative clauses with possessive relative phrases, phrases which contain the relative pronoun hvis.

1. (7)

   

Footnote ¹³Footnote ¹⁴

In (8) the free relative phrases are followed by a bound relative clause with a prepositional relative phrase.

1. (8)

   

Footnote ¹⁵

And finally, in (9) we find examples of hvor som helst ‘wherever’, followed by bound relative clauses with the relative pronoun hvor ‘where’, which is equivalent to på hvilket ‘on which’.

1. (9)

   

Footnote ¹⁶Footnote ¹⁷

The occurrence of a following bound relative clause is restricted to cases involving pied piping. This also means that the free relative construction is not an ordinary relative construction, with an ordinary relative clause. If it were, we would not expect this restriction, and examples like those in (10) would be expected to be well-formed. Such examples are not well-formed.Footnote ¹⁸

1. (10)

   

3. BACKGROUND

Danish free relative constructions have been discussed in the Danish literature under various headings. Mikkelsen (Reference Mikkelsen1911:509) discusses uegentlige (improper) relative clauses which are defined as relative clauses with no explicit correlate in the main clause. Diderichsen (Reference Diderichsen1957:210) talks about a type of relative clause where the correlate is only expressed in the relative clause. Hansen (Reference Hansen1967:200) talks about indefinite relative clauses where the free relative pronoun does not refer back to a constituent in the preceding clause. Hansen & Heltoft (Reference Hansen and Lars2011:1537) talk about almene (general) relative clauses which lack an explicit correlate in the main clause, i.e. they lack a nominal head. What all these accounts have in common is that they assume that the entire free relative construction is a relative clause.

The above Danish analyses are in line with the traditional analysis of free relatives within the generative paradigm, where free relatives are also assumed to be clauses, projected to NPs, where the wh-phrase is extracted to clause-initial position, see e.g. Chomsky (Reference Chomsky, Anderson and Kiparsky1973). Example (11) shows this analysis (from Bresnan & Grimshaw Reference Bresnan and Grimshaw1978:331).

1. (11)

   

Example (12) shows the alternative structure, assumed by Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978).Footnote ¹⁹ Importantly, in a free relative, the wh-phrase is a base-generated head rather than extracted from the sister clause, see Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978:331).

1. (12)

   

The process by which a free relative is derived is called Controlled Pro Deletion, (Bresnan & Grimshaw Reference Bresnan and Grimshaw1978:368–378). Controlled Pro Deletion deletes the terminal string of a pro-category and co-indexes the pro-category with the antecedent free relative phrase.

The process is illustrated in (13) and (14), where the terminal string △ of an NP pro-category is deleted and co-indexation between the pro-category and the antecedent free relative phrase is established (Bresnan & Grimshaw Reference Bresnan and Grimshaw1978:369).

1. (13)

   

2. (14)

   

According to Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978), the analysis of free relatives as base-generated wh-phrase–headed constructions is confirmed by a number of grammatical phenomena. Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978) show that assuming the wh-phrase to be the head of a free relative explains the behaviour of English free relatives with respect to e.g. the matching effect, number agreement, the internal NP over S constraint, the independent generation of non-specific free relatives and PP pied piping, see also Bjerre (Reference Bjerre and Müller2012).

The independent generation of non-specific free relatives is the only argument concerning the part of the hypothesis that the free relative phrase is base-generated, see Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978:339–342). The argument is that a non-specific free relative pronoun may occur without a following sister clause, as in (15), see Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978: 339–340).

1. (15)

   1. a. She wrote whenever possible.

   2. b. She'll go wherever possible (to promote her cause).

   3. c. She vowed to do whatever possible to vindicate herself.

In (15) the free relative occurs with an adjetive. Bresnan & Grimshaw argue that such constructions are not derived transformationally from full clauses. This would require an explicit listing of the adjectives which undergo the transformation, as illustrated in (16), (Bresnan & Grimshaw Reference Bresnan and Grimshaw1978:340).

1. (16)

   1. a. She wrote whenever it was possible. →

   2. b. She wrote whenever possible.

2. (17)

   1. a. She didn't write whenever it was impossible . →

   2. b. *She didn't write whenever impossible.

When a non-specific wh-phrase occurs without a sister clause, there is no sister clause from which it can have been extracted and consequently the wh-phrase is a base-generated head.

In Danish, non-specific free relative pronouns also commonly occur without a dependent clause, as in (18).

1. (18)

   

Footnote ²⁰Footnote ²¹ And in Danish this is also restricted to non-specific free relative constructions, as shown in (19).

1. (19)

   

The dependent clause after the specific free relative is obligatory. The optionality/obligatoriness of the dependent clause is determined by the antecedent ‘fused’ into the relative pronoun. This is illustrated in (20).

1. (20)

   

The relative clause after the definite antecedent with the definite article is obligatory, whereas the relative clause after the universally quantified noun is optional.

4. HPSG WH-HEAD ANALYSES OF FREE RELATIVES

As explained in Bjerre (Reference Bjerre and Müller2012), a number of HPSG accounts of free relatives have adopted the wh-head analysis, see e.g. Kim (Reference Kim2001), Wright & Kathol (Reference Wright and Kathol2003), Kubota (Reference Kubota2003), Taghvaipour (Reference Taghvaipour and Müller2005) and Borsley (Reference Borsley2008).Footnote ²² However, these analyses do not adopt the idea that the wh-phrase is base-generated and not extracted from its sister clause, as in Bresnan & Grimshaw (Reference Bresnan and Grimshaw1978); rather, in HPSG terms, they assume that there is a filler–gap relation between the wh-phrase and a gap in the sister clause of the wh-phrase. The structure in (21), slightly modified from Kim (Reference Kim2001:44), illustrates the structure of free relatives that is assumed by these analyses.

1. (21)

   

As this example shows, the constituent is an NP. At the same time, though, the constituent is a head–filler structure which contains a filler free relative pronoun and a sister clause with a missing constituent.Footnote ²³ The NP does not miss a constituent, as the filler free relative pronoun cancels the missing constituent. The structure implies that the free relative pronoun has been extracted from its sister clause in the sense that its canonical position is in the sister clause. There is no implication that actual movement is involved.

5. OUR ANALYSIS

The data in Section 2 suggests that the structure of Danish free relatives does not involve a clause from which the free relative pronoun has been extracted. We propose that the gap in the sister clause in a free relative is cancelled off before forming a constituent with the free relative pronoun phrase. The wh-phrase is the head of an NP and the sister clause is a bound relative clause, albeit with the restriction mentioned in Section 2 that the bound relative clause involves pied piping.

Importantly, we need to look at the structure of the free relative constructions with bound relative clauses. The structure proposed for (7a) above is shown in (22).

1. (22)

   

The gap in the relative clause is filled by the phrase containing the relative pronoun hvis ‘whose’. There is no gap for the non-specific free relative pronoun phrase to fill in the modifying sister clause.

The diagram in (23) shows the structure for the free relative hvem som helst der citerer Dickens ‘whoever there cites Dickens’.

1. (23)

   

In this example, the gap is cancelled off by a construction, as the relative clause does not contain a relative pronoun filler.

The structure we propose is similar to the structure of an ordinary relative construction like (24), which shows the structure of the relative construction artikler der citerer det givne værk (literally: ‘articles there cite the given work’), and, as can be seen, it is similar to the structure in (23).

1. (24)

   

Supporting evidence for our proposed analysis of free relatives comes from extraposition data. Ordinary relative constructions allow extraposition of the relative clause out of the NP. This is illustrated in (25) where the relative clauses are separated from their head noun by a verbal particle.

1. (25)

   

Footnote ²⁴Footnote ²⁵Footnote ²⁶ In non-specific free relatives, the free relative pronoun can also be separated from its sister clause by a verbal particle, as shown in (26).

1. (26)

   

Footnote ²⁷Footnote ²⁸Footnote ²⁹ The extraposition in (26) is accounted for on the assumption that the free relative pronoun's sister clause is a relative clause in itself without a missing constituent modifying the free relative phrase head. On this assumption, we are dealing with relative clause extraposition. If, on the other hand, the free relative construction itself is a clause, the extraposition behaviour is not readily accounted for. Then the free relative construction would have a structure similar to e.g. an interrogative clause. As (27) shows, interrogative clauses do not allow extraposition of the sister clause of the interrogative pronoun.

1. (27)

   

Footnote ³⁰

6. FORMALIZATION

In this section we will show the formalization of our analysis. We will use the free relative construction in example in (28) to illustrate the formalization.

1. (28)

   

The sentence has the structure shown in (29).

1. (29)

   

The formalization is based on Ginzburg & Sag (Reference Ginzburg and Sag2000) and Sag (Reference Sag1997) using a gap-ss type to represent gaps, the Argument Realization Principle to remove gap-ss arguments from the valence lists, the SLASH-Amalgamation Constraint to determine the SLASH value of a word, the Generalized Head Feature Principle to propagate the SLASH value, and a head–filler phrase or constructional gap-binding to finally bind off the gap. Our formalization also uses the expl(etive)-ss type and the revised Argument Realization Principle for Danish proposed in Bjerre (Reference Bjerre, Pedersen, Nĕspore and Skadiṇa2011a, Reference Bjerre and Müllerb, Reference Bjerre and Müller2013).

In our analysis, expletives in relative clauses appear on the ARG -ST list and are put in the SLASH set like ordinary gaps to be bound off at some point in the structure. (30) shows the hierarchy of synsem types assumed in this analysis, see Bjerre (Reference Bjerre and Müller2011b:281).

1. (30)

   

The canon-ss type is subtyped into an expl(etive)-ss and a non-expl(etive)-ss, the former representing the expletive occurring in subject position. In (31) and (32), the constraints on the gap-ss (see Sag Reference Sag1997:446; Ginzburg & Sag Reference Ginzburg and Sag2000:170), and the expl-ss (Bjerre Reference Bjerre and Müller2011b:282) are shown.

1. (31)

   

2. (32)

   

The gap-ss has neither syntactic nor semantic content of its own. The entire LOCAL value is put in its SLASH set. The expl-ss, on the other hand, has syntactic content of its own, i.e. the value of HEAD is the category expl(etive), and only its CONTENT value is structure-shared with an otherwise underspecified element in its SLASH set.Footnote ³¹ The constraint for the verb smager ‘tastes’, as it occurs in (29) above, is given in (33).

1. (33)

   

The verb takes two arguments, an expletive and an AdvP.

In (34), the SLASH-Amalgamation Constraint from Ginzburg & Sag (Reference Ginzburg and Sag2000:169) is shown. The constraint determines the SLASH value of a head word by amalgamating all the SLASH values of its arguments.

1. (34)

   

Applying the SLASH-Amalgamation Constraint to the entry in (33) above results in the entry in (35), where the SLASH set for smager ‘tastes’ is defined.

1. (35)

   

The Argument Realization Principle for Danish in (36) (see Bjerre Reference Bjerre and Müller2011b:282) is used to map the lexical arguments on the ARG-ST list on to the valence lists representing syntactic functions.

1. (36)

   

The valence lists, i.e. the SUBJ and COMPS lists, are used to represent syntactic functions whereas the ARG-ST list represents argument structure at the lexical level ignoring syntactic functions. The ⊖ symbol is used for contained list difference. The expression L1 ⊖ L2 is the list resulting from removing the elements in L2 from L1, see Ginzburg & Sag (Reference Ginzburg and Sag2000:170).

The elements on the ARG-ST list are initially underspecified with respect to whether they are canon-ss or gap-ss types. If an argument is resolved to a gap-ss, the principle excludes it from both the SUBJ(JECT) list and the COMP(LEMENT)S list, i.e. we analyze subject gaps as being extracted. On the other hand, if an argument is resolved to an expl-ss, it appears on the SUBJ list, even though it also gives rise to an element in the SLASH set.

Applying the Argument Realization Principle for Danish to our entry in (35) gives rise to the entry in (37).

1. (37)

   

The expletive on the ARG-ST list is structure-shared with an element on the SUBJ list and the AdvP appearing on the ARG-ST list is structure-shared with an element on the COMPS list. In this example none of the elements on the ARG-ST list is removed from the valence lists.

In (38) the verb smager ‘tastes’ combines with the AdvP complement godt ‘well’.

1. (38)

   

The element on the COMPS list is cancelled off and the mother's COMPS list is empty. The other features, including the SLASH value, are propagated by the Generalized Head Feature Principle from Ginzburg & Sag (Reference Ginzburg and Sag2000:33). The constraint is a default constraint and all the features and their values in SYNSEM are propagated by default. The default rule can be overridden by other constraints, e.g. a constraint binding off an element from the SLASH set, or, as in this example, a constraint cancelling off the element on the mother's COMPS list. The Generalized Head Feature Principle is shown in (39).

1. (39)

   

The VP is then combined with the expletive subject der ‘there’ to form a clause, as shown in (40).

1. (40)

   

The element on the SUBJ list is cancelled off and the mother's SUBJ list is empty. The Generalized Head Feature Principle ensures that the SLASH value from the entry for smager ‘tastes’ is present at clause level

As explained in Section 5 above, we do not assume a filler–gap relation between the free relative phrase and a gap in its sister clause. The SLASH element due to the gap filled by an expletive is not bound off by the free relative phrase in a head–filler-phrase. Instead we assume that the SLASH element is bound off constructionally by the constraint in (41), see Sag (Reference Sag1997:36).

1. (41)

   

Constructional gap-binding is used when there is no relative pronoun to bind off the gap. If, on the other hand, we had an example like (22), the gap would be bound off in a subtype of a wh-rel(ative)-cl(ause) constraining the filler to be a pied piped phrase. Applying the constraint in (41) to our clause in (40) we get a relative clause as shown in (42).

1. (42)

   

As can be seen, the top node now has the MOD feature characterizing a modifying relative clause. Further, the SLASH set is empty, as the element in it has been bound off.

Finally, the relative clause combines with the free relative phrase to form an NP. The resulting construction is shown in (43).

1. (43)

   

The index on the slashed element and the modified element are structure-shared, and this index is again structure-shared with the index on the wh-phrase. In this way the relation between the gap and the free relative pronoun is established. That the wh-phrase is the head of the construction can be seen by the structure sharing between the HEAD values of the wh-phrase and the top node.

7. IMPLICATION OF OUR FORMAL ANALYSIS

In this section we will explore one consequence of our formal analysis. Our formal analysis does not allow for free relative constructions involving PP pied piping. In Section 2 we argued that such free relative constructions are not allowed. There are, however, examples of free relatives which apparently involve such pied piping. Müller (Reference Müller1999) provides the German equivalent of the Danish example in (44) which apparently involves PP pied piping.

1. (44)

   

Such examples are restricted, though, and it should be noted that categorial matching of the PPs is not enough to achieve well-formedness. Note that the example in (45) is not well-formed.

1. (45)

   

In such examples the verbs in the main clause and the relative clause must be identical. Examples without the apparent pied piping are of course possible, as shown in (46).

1. (46)

   

In analyses allowing pied piping of the free relative pronoun phrase, the example in (44) becomes structurally ambiguous, as shown in (47), where (47b) is the pied piping analysis.

1. (47)

   

In our analysis, (47b), involving PP pied piping, is not possible. The gap in the sister clause is bound off by the constraint in (41) which can only bind off NP gaps. As the structural ambiguity does not reflect a semantic ambiguity, we believe that this limitation of our formalization is desirable. The analysis provided for the free relative construction in (47a) is shown in (48).

1. (48)