1. Introduction
In many languages, traditional grammatical descriptions recognize a class of auxiliaries distinct from main verbs. They are distinguished by both language-independent and language-specific characteristics. By definition, auxiliaries can co-occur with a main verb, and they do not contribute to the thematic structure of the clause (Schachter 1985:41–44, Payne 1997:84f.). In English, auxiliaries raise to T, to the left of negation and left-edge adverbs, and they require the verbal element immediately following them to appear in a special inflectional form (participial, infinitive, and so on) (Jespersen 1931:11). In one early generative line of investigation, started by Ross (1969) and continued by Pollock (1989) and others, this distinction was argued to be unnecessary. Auxiliaries were simply verbs that selected another verb phrase as complement. The past fifteen years, however, have witnessed a resurgence of the more traditional view, according to which auxiliaries comprise a conceptual category distinct from main verbs.
To begin with, the generative conception of clause structure has expanded dramatically to include a variety of functional heads. The v head, in particular, introduces the external argument so that the “main verb” is actually a composite of a functional v head and a lexical V head (Hale & Keyser 1993b, Kratzer 1996). This more articulated clausal structure presents a problem for the analysis that treats auxiliaries as verbs that select a verb-phrase complement. The question is whether auxiliaries have both a lexical and a functional component, just like main verbs do—that is, whether auxiliaries have their own v, or can auxiliaries simply be analyzed as functional projections located above the verb?
The current consensus is that auxiliaries should be treated as functional heads. In Adger’s (2003:155–203) implementation of this view, the English auxiliaries fall into four categories: modals (Mod), the perfect have (Perf), the progressive be (Prog), and the passive be (Pass). All four functional heads are located above v and below T. Under this view, while auxiliaries are internally diverse, they are still distinct from main verbs because they occur above both v and V.Footnote 1
While in English the division between auxiliaries and main verbs seems relatively clear cut, making the same distinction in Danish is not as easy. On the basis of language-independent properties, Danish does have a class of auxiliaries that includes a variety of modals, the perfect have ‘have’, the perfect være ‘be’, and the passive blive ‘become’. These verbal forms co-occur with main verbs and do not affect the assignment of thematic roles.Footnote 2 However, the Danish auxiliaries differ from their English counterparts in a number of ways. In root clauses, Danish has verb-second word order: The highest verbal element, whether it is an auxiliary or a main verb, raises to T and then to C, where it follows the first major sentence constituent (Vikner 1995, among others). In most embedded clauses—including adverbial clauses, clausal subjects, clausal complements to non-bridge verbs (Iatridou & Kroch 1992), relative clauses, and clausal complements of nouns—all verbal elements stay in situ, where they surface below negation and other left-edge elements.Footnote 3 Consider, then, what the surface position of a miscellaneous verbal head X located between T and v would be. In matrix clauses, X would be indistinguishable on the surface from T since it would raise and head-adjoin to it, thus occurring to the left of negation and other left-edge elements (see 1a). By contrast, in embedded contexts, X would be in-distinguishable from v, which in Danish is null, or from V (see 1b).
- (1)
a. Matrix clause
b. Embedded clause
We are concerned here with one verbal element in Danish, g⊘re. Following traditional grammars (Diderichsen 1966:63, Hansen 1967:69, Allan et al. 1995:143f., 159, 511f.), we assume that g⊘re is an auxiliary and propose that it is located above v but below T in the extended verbal projection—precisely where X is located in 1—despite the issues this raises.Footnote 4 In root clauses, g⊘re moves to, and is indistinguishable from, T. In embedded clauses, g⊘re is especially hard to tell apart from v because it has a restricted syntactic distribution. G⊘re only appears when the verb phrase has been topicalized, as in 2a, elided, as in 2b, or realized as a verbal proform det, as in 2c.Footnote 5
- (2)
a.
b.
c.
In each construction in 2, the lexical verb V does not appear in its canonical position—it is either fronted, deleted, or replaced by a proform. Therefore, g⊘re could very plausibly originate very low in the clause—potentially as a v. We argue, nonetheless, that the only way one can account for the full range of g⊘re’s properties is to analyze it as an auxiliary that appears above v but below T.
Our argument proceeds as follows. In section 2 we set aside various alternative analyses of g⊘re. First, we show that g⊘re is not simply a realization of T since it can occur to the left of negation and left-edge elements in embedded clauses (section 2.1). Then, we distinguish between g⊘re in verb phrase ellipsis, topicalization, and pronominalization on the one hand, and g⊘re used as a main verb on the other hand (section 2.2). Nor can g⊘re plausibly realize v (section 2.3), an approach championed by Platzack (to appear). On the one hand, g⊘re has none of the characteristics that overt v’s have in other languages. On the other hand, as we show in section 3, g⊘re has a number of properties characteristic of auxiliaries. The proper analysis, then, is one where g⊘re is an auxiliary and occurs in the extended verbal projection between T and v. Additional evidence for the auxiliary status of g⊘re comes from occurrences of nonfinite g⊘re (section 3.1), co-occurrence restrictions with other auxiliaries (section 3.2), and the lack of ellipsis with infinitival g⊘re (section 3.3). As an auxiliary, however, it is defective—it has the restricted range of distribution illustrated in 1. We propose that, as the lowest auxiliary of the clause, g⊘re takes vP as its complement, though it subcategorizes for only certain types of vPs, namely, those that are pronominal. In section 4, we show how this leads to g⊘re appearing only in the three environments identified above. Section 5 is a conclusion.
2. The Category of G⊘re
As stated in the introduction above, depending on the syntactic context g⊘re could be treated as any one of a number of verbal projections. Some of these analyses are relatively easy to set aside. First, we show that while the English pleonastic do that appears in negation, verum focus, inversion, and other contexts is often treated as a last resort pronunciation of T (Chomsky 1957:62–69, Halle & Marantz 1993:134–137, Bobaljik 1995:63–78), Danish g⊘re cannot be analyzed this way. Second, we argue that g⊘re cannot be analyzed simply as an instance of V. While there is a homophonous main verb in Danish, various tests show that it is clearly distinct from the g⊘re we are interested in. Furthermore, since under a contemporary view of clause structure the main verb corresponds to the V-v complex, the same tests indicate that g⊘re cannot be analyzed as v. This leaves the most reasonable analysis of g⊘re as an auxiliary.
2.1. G⊘re is Not a T
As schematized in 1, in main clauses g⊘re raises to T, just like other finite verbal elements. In a previous analysis, we took g⊘re to be inserted into T in a last-resort operation that parallels English do-support (Houser et al. 2006). Embedded clauses show that g⊘re must originate lower in the extended verbal projection. In the relative clause in 3a, the present tense form g⊘r appears below a left-edge adverb.Footnote 6 In 3b, g⊘re appears to the right of negation in the protasis of a conditional. Finally, in 3c, g⊘re surfaces inside a sentential subject, below the left-edge adverb stadig ‘still’.Footnote 7
- (3)
a.
b.
c.
If g⊘re were inserted directly into T, then it would occur to the left of negation and other left-edge elements. However, this order is ungrammatical:
(4)
However, if one assumes that g⊘re originates in the verbal projection below T and then follows the standard analysis of embedded clauses, according to which T lowers onto the highest verbal element in the clause, the correct ordering of g⊘re with respect to left-edge elements falls out naturally:
(5)
The question is whether g⊘re could originate in T, then lower onto another verbal projection, and thus appear to the right of negation and left-edge adverbs. We discuss three reasons why this could not be the case. First, there would be no reason for g⊘re to lower. As a bound tense morpheme, T lowers to join a morphological host (Chomsky 1957:38–42). However, g⊘re is a free morpheme and as such does not need a host. Second, as we discuss in section 4, g⊘re is only possible when the vP has been topicalized, elided, or replaced by a proform det. If g⊘re were in T, then there would be no verbal head for it to lower onto. Third, if g⊘re were in T one would never expect it to occur below any other verbal forms. Contrary to this expectation, the past participle form of g⊘re (gjort) occurs as the complement of perfect have, as in 6.
(6)
Such instances of nonfinite g⊘re suggest that it cannot be analyzed as T. Moreover, as we discuss in section 3, they provide crucial evidence that g⊘re is an auxiliary.
2.2. G⊘re is Not a Main Verb
If g⊘re is not a realization of T, it could originate lower in the clause, as a V. Danish does have a main verb g⊘re, which, like the English main verb do, is transitive and has a highly schematic meaning that does not indicate the type of action undertaken on its own.
- (7)
a.
b.
c.
A number of distributional and interpretive properties distinguish the main verb g⊘re from the auxiliary g⊘re that we are interested in here (see Jäger 2006:83f.). Specifically, only the auxiliary g⊘re exhibits the following five properties:
(i) It can co-occur with lexical verbs in the same clause.
(ii) It does not contribute to the conceptual structure of the clause.
(iii) It does not introduce its own event.
(iv) It does not add, remove, or modify the thematic roles.
(v) It does not add, remove, or modify the grammatical relations.
Starting with property (i), g⊘re can co-occur with lexical verbs, such as vaske ‘to wash’ or even with main verb g⊘re, as shown in 8:
(8)
Since the vP g⊘re ondt ‘to hurt’ (literally, ‘to do evil’) is fronted, g⊘re, which occurs in second position, must be of a category other than V.
The properties in (ii)–(v) are illustrated below: 9b is the topicalization sentence, while 9a is the corresponding basic sentence with the vP in its original position.
- (9)
a.
b.
As the examples show, g⊘re does not add anything to the conceptual structure of the sentence since the two sentences in 9 are truth-conditionally equivalent. It does not contribute its own event to the meaning of 9b either: just like 9a, 9b is construed as describing a single event of car washing. Finally, g⊘re does not affect thematic roles or grammatical relations in the topicalization sentence. In both 9a and 9b, han ‘he’ is the agent and subject, while bilen ‘the car’ is the patient and direct object. In all these respects g⊘re does not behave like a V.
2.3. G⊘re Is Not a v
Finally, g⊘re could potentially be analyzed as an instance of v, as proposed by Platzack (to appear).Footnote 8 He treats g⊘re in Danish (as well as its cognates in the other Scandinavian languages) as the realization of v whenever g⊘re is not adjoined to V.Footnote 9 This happens when the VP has been elided, topicalized, or replaced by a proform. Thus, under Platzack’s analysis the ellipsis sentence in 10 (repeated from 2b) has the structure in 11.
(10)
(11)
This analysis makes correct predictions regarding the distribution of g⊘re in embedded clauses. As the realization of v, g⊘re is expected to appear to the right of left-edge elements. However, if g⊘re is v it is expected to behave like other members of this category in other respects as well. Looking solely at Danish, this prediction is difficult to test since there are no other overt members of this category. Thus, we examine verbal elements in other languages that have been argued to instantiate the category of v.
The primary representatives of this category are light verbs, a term originally coined by Jespersen (1946:117) to describe English verb-NP combinations such as take a rest or have a nap. According to Butt (2003), in light verb constructions the core semantic content of the clausal predicate is provided by something other than the light verb, though the light verb contributes to the argument or event structure of the predicate.Footnote 10 This is exactly the contribution of v to the meaning of the predicate (Hale & Keyser 1993b, Kratzer 1996). Therefore, light verbs are perfect candidates to realize v overtly, as they indeed have been argued to do in a number of different languages, such as French and Italian (Folli & Harley 2007), Hindi-Urdu (Butt & Ramchand 2005), Persian (Megerdoomian 2002, Folli et al. 2005), and Yiddish (Diesing 1998).
Persian is one of the languages that show this most clearly. It has an inventory of light verbs that, together with a nonverbal element (a noun, an adjective, or a PP), form the vast majority of predicates in this language.Footnote 11 Light verbs form pairs that participate in argument structure alternations. Two of these, zadan ‘to hit’ and khordan ‘to eat’, create complex predicates that differ only in whether or not they have an external argument. The complex predicate laqat zadan ‘to kick’ is transitive, while laqat khordan ‘to get kicked’ is unaccusative.
- (12)
a.
b.
c.
Only when zadan is present can the predicate have an external argument, as illustrated by the ungrammaticality of 12c. This leads Megerdoomian (2002) and Folli et al. (2005) to conclude that the light verbs, including zadan, are v’s responsible for introducing the external argument.
At first glance, g⊘re can be analyzed as a light verb, just like Persian zadan or khordan (see Platzack to appear). As Toosarvandani (2009) argues, light verbs in Persian are able to license ellipsis of their complement that contains the nonverbal element and internal argument, as shown in 13.
(13)
However, g⊘re differs from Persian light verbs in two crucial respects. First, following the view in this paper, if g⊘re were a v, it would contribute to the content of the predicate. As we already showed in section 2, however, this is not the case. Second, g⊘re affects neither the predicate’s theta-assigning properties nor the grammatical relations within the sentence. In contrast, the Persian light verbs zadan ‘hit’ and khordan ‘eat’ require and forbid an external argument, respectively.
These considerations alone do not rule out entirely the analysis of g⊘re as v. It is possible that g⊘re does not show the same properties as other light verbs. Since in Danish all members of the v category are either null or realized as g⊘re paradigmatic alternations of the type found in Persian are not perceptible on surface. That is, in order to maintain a v analysis one would have to posit a collection of homophonous v heads g⊘re. One such v would introduce an external argument in transitive and unergative predicates, while another would occur in unaccusative predicates. Creating this type of radical homophony does not strike us as particularly insightful, but it is not necessary to rely solely on such considerations of parsimony to exclude this analysis. In the next section, we examine nonfinite forms of g⊘re and argue that it should be analyzed as an auxiliary and not a light verb of category v.
3. G⊘re is an Auxiliary
We propose that g⊘re is an auxiliary that heads a functional projection located above v but below T. The sentence in 10 has, under our analysis, the structure in 14.
(14)
This analysis derives the correct word order with respect to the left-edge elements. Negation and left-edge adverbs are adjoined to the complement of T, which in 14 is the AuxP headed by g⊘re. Thus, in embedded clauses g⊘re appears to the right of these elements.
It follows from our analysis that when there are no other auxiliaries, the presence of g⊘re in the extended verbal projection is obligatory. We assume that finite inflection is associated with its own terminal node, T. When the vP is elided, topicalized, or pronominalized there is no main verb that can function as a morphological host for this inflection. If g⊘re (or some other auxiliary) is not present in the numeration, then the finite inflection does not have a host, which violates the Stray Affix Filter—the requirement that a bound morpheme attach to a phonological host (Baker 1988:140).
If g⊘re is an Aux, it should behave like other members of that category. As we discussed in section 1, there are language-specific as well as language-independent criteria for determining if a verb is an auxiliary.Footnote 12 In particular, main verbs in Danish cannot appear in tag questions, as shown by the ungrammaticality of 15a, while Danish auxiliaries can, as shown in 15b. If g⊘re is an auxiliary the prediction is that it should be able to appear in tag questions as well. This prediction is borne out, as shown in 15c.
- (15)
a.
b.
c.
Another language-specific property of auxiliaries in Danish is that they must be followed by either an infinitival form (in case of modal auxiliaries) or a participial (in case of perfect have and være, passive blive). Of course, no verb follows g⊘re when the vP is elided or pronominalized. However, when the vP is topicalized (as in 2a repeated here as 16), the main verb appears in the infinitive.
(16)
The language-independent properties of auxiliaries are some of the same ones we used in section 2 to show that g⊘re, when it appears in verb phrase ellipsis, pronominalization, and topicalization contexts, is not a main verb. Of course, auxiliaries can co-occur with a main verb in the same clause (though g⊘re only does so when the verb phrase has been topicalized). In addition, as L⊘drup (1996) shows, they do not impose any semantic restrictions on the subject. That is to say, auxiliaries do not contribute to the thematic structure of the clause.Footnote 13
Until now, we have discussed only finite g⊘re. However, as we mentioned in section 2, there are also occurrences of nonfinite g⊘re. In the rest of this section, we show that nonfinite g⊘re provides two arguments in support of our analysis of this verb as an auxiliary. First, as discussed in section 3.2, nonfinite g⊘re does not occur as the complement of the perfect auxiliary være or the passive auxiliary blive. This distributional gap can be accounted for only if g⊘re is analyzed as an auxiliary. Second, as discussed in section 3.3, infinitival g⊘re does not license vP ellipsis. This is expected if g⊘re is an auxiliary, but not if it is a v.
3.1. Non-Finite G⊘re
When g⊘re combines with the perfect auxiliary have it appears in the participial form gjort, as shown in 17. When g⊘re follows a modal, it appears in the infinitival form g⊘re, as shown in 18.
- (17)
a.
b.
c.
- (18)
a.
b.
c.
Nonfinite auxiliary g⊘re satisfies the same five criteria described in section 2 that finite g⊘re does. For the first criterion, the participial form gjort co-occurs with the main verb dominere ‘dominate’ in 17a, and the infinitival form g⊘re co-occurs with the main verb unders⊘ge ‘investigate’ in 18a. Satisfaction of the remaining four criteria can be established by comparing 17c and the parallel sentence in 19, which does not contain participial gjort and does not involve ellipsis.
(19)
Nonfinite g⊘re does not contribute to the conceptual structure of a sentence (criterion ii): 17c and 19 have identical truth conditions. It also does not introduce its own event (criterion iii) since the relative clause describes a single event in both sentences, that of Teddy staying at the hotel in question. Finally, nonfinite g⊘re does not affect the thematic or grammatical roles of the clause it appears in (criteria iv and v). In both 17c and 19, Teddy is the subject and agent, and samme hotel ‘the same hotel’ is the object of the preposition pâ and it specifies a location. We can thus be certain that we are not dealing with main verb g⊘re here.
Like finite g⊘re, nonfinite g⊘re cannot appear with a canonically realized vP. Compare 17a that contains a topicalized vP and nonfinite gjort with the ungrammatical 20a, where the vP occurs in situ. Similarly, compare 17c, where nonfinite gjort occurs with vP ellipsis to the ungrammatical 20b with no ellipsis. Finally, compare 18b, where non-finite g⊘r co-occurs with the verbal proform det, to the ungrammatical 20c.
- (20)
a.
b.
c.
There is, however, one significant difference between finite and nonfinite g⊘re. Finite g⊘re is obligatory (as is any finite verb in Danish)—in the sense that leaving finite g⊘re out results in ungrammaticality—while nonfinite g⊘re is not (though, for relevant discussion, see Ørsnes, to appear). As shown in 21, ellipsis is felicitous with participial gjort (the attested form) or without it.
(21)
Participial gjort is optional in 17a, and infinitival g⊘re is optional in 18c (examples repeated below as 22a and 22b, respectively). In our analysis of g⊘re as an auxiliary, this follows from basic properties of all auxiliaries.
- (22)
a.
b.
Like perfect have or any other auxiliary, g⊘re is not obligatorily present in the extended verbal projection. There are numerations that converge without it. When nonfinite g⊘re is present, its sister, vP, is topicalized, elided, or pronominalized. Since the other auxiliaries license these constructions as well, when nonfinite g⊘re is not present, they take the vP directly as complement.
3.2. Subcategorizing for G⊘re
The first argument in favor of our analysis of g⊘re as an auxiliary involves the perfect auxiliaries have ‘have’ and være ‘be’. Most verbs require have as an auxiliary, but a sizable minority appears with være, including falde ‘fall’, komme sig ‘recover’, and smelte ‘melt’ (see Vikner & Sprouse 1988 and Bjerre & Bjerre 2007 for discussion). As a starting point, we should point out that g⊘re is not inherently incompatible with main verbs that require være. Verb phrases with være can serve as antecedents of pronominalized and elided vPs introduced by finite g⊘re, as shown in 23.
- (23)
a.
b.
In 23a and 23b, finite g⊘re licenses pronominalization and ellipsis, respectively. Nonfinite g⊘re is also found with an antecedent that contains være in the perfect.
- (24)
a.
b.
Even though g⊘re is compatible with verbs that require være in the perfect, it cannot itself co-occur with perfect være, as shown in 25. In contrast, g⊘re co-occurs quite felicitously with perfect have, as shown in 17 and 21 above.
- (25)
a.
b.
Similarly, g⊘re cannot co-occur with the passive auxiliary blive, as shown in 26a.Footnote 14 Again, this restriction is not due to a more general incompatibility with a passive complement since g⊘re can have a passive antecedent, as shown in 24b, where the antecedent is the passive-participle fulgt ‘followed’.
- (26)
a.
b.
Consider now how these restrictions on nonfinite g⊘re are accounted for under the auxiliary analysis. As an auxiliary, g⊘re is generally optional. When it occurs it is selected for by a higher head, which is either T (yielding finite g⊘re) or another auxiliary (yielding nonfinite g⊘re). The same is true for any regular auxiliary. It is a complement of either T or another auxiliary. While T can take any auxiliary as its complement, auxiliaries are typically more selective. Thus, while perfect have does not allow complements headed by the passive blive, perfect være does. While perfect være does not allow modal complements, perfect have does. We suggest that the restrictions on nonfinite g⊘re should be accounted for along the same lines: G⊘re is incompatible with perfect være because the latter does not subcategorize for g⊘re. Similarly, nonfinite g⊘re is incompatible with passive blive because blive does not sub-catagorize for g⊘re. In fact, blive cannot take auxiliaries as complements at all. These restrictions are summarized below:
- (27)
a. modal < perfect have < modal < g⊘re
b. modal < perfect være < passive blive
G⊘re belongs only to the auxiliary hierarchy in 27a since it is subcategorized for by modals and perfect have, but not by være or blive. There might be explanations for these subcategorization restrictions, but what is important here is that g⊘re patterns with other auxiliaries in that it is selected by some, but not all auxiliary verbs. G⊘re itself does not allow for an auxiliary complement, but that does not disqualify it from being an auxiliary. For example, the passive auxiliary blive does not take an auxiliary complement either.
The dependence of nonfinite g⊘re on the auxiliary immediately above is entirely unexpected under Platzack’s analysis. From the outside, there is no syntactic difference between a vP in which V has raised to v on the one hand, and one in which V fails to raise and v is realized as g⊘re, on the other hand. For Platzack, this is an alternation governed entirely by the featural properties of V, and v itself is the same in both cases. Since the two types of vP are indistinguishable from higher auxiliaries, any auxiliary should be able to select a vP in which V does not raise to v and v is pronounced as g⊘re. This is a prediction that Platzack himself acknowledges (p. 13): “If the support verb is a spelled-out v, we expect it to occur after all auxiliaries.” In particular, one would expect it to follow perfect være and passive blive, contrary to fact.Footnote 15
Under the v analysis, to account for the inability of g⊘re to co-occur with perfect være and passive blive, one might say that these auxiliaries select the VP as their complement rather than the vP. There would be no v in structures 25 and 26a above, and hence no potential position for g⊘re. Though note Platzack commits himself explicitly (p. 9) to v being present in all clauses. It is clear, however, that some complements of perfect være and passive blive do contain a v. As shown in 23 and 26b, finite g⊘re is compatible with these verbs, and clauses with finite and nonfinite g⊘re have, by hypothesis, the exact same structure. To fully analyze finite and nonfinite g⊘re one would have to posit that v is optional in the structure. However, while some auxiliaries (modals and perfect have) would take either vP or VP as their complement, perfect være and passive blive would only take VP. This move seems entirely ad hoc to us as it vitiates any motivation for including v in the extended verbal projection in the first place. If v is optional what precisely does it do?
From the hierarchy in 27a one can tell that it is hard to distinguish between the two analyses of g⊘re, that is, the v analysis and the auxiliary analysis. On the v analysis, g⊘re appears immediately below the lowest auxiliary. On the auxiliary analysis, g⊘re is the lowest auxiliary. We conclude, nonetheless, that it is possible to differentiate between the two approaches. The ban on the co-occurrence of nonfinite g⊘re with perfect være and passive blive mimics gaps in the distribution of other auxiliaries. This similarity is captured by the auxiliary analysis that accounts for all such gaps in terms of subcategorization.
3.3. No Ellipsis with Infinitival G⊘re
The difference between the auxiliary analysis and the v analysis becomes relevant in another area as well. When g⊘re is in its infinitival form it does not license verb phrase ellipsis, as shown in 28a–c. However, verb phrase pronominalization and topicalization are both possible, as shown in 17 repeated below as 29. The other nonfinite form of g⊘re, past participial gjort, is not subject to this restriction, as shown by the grammatical ellipsis example in 28d, and 16c and 20 repeated below as 30 and 31, respectively.
- (28)
a.
b.
c.
d.
- (29)
a.
b.
c.
(30)
(31)
In this respect, infinitival g⊘re patterns with regular auxiliaries. The infinitival forms of perfect have and passive blive are unable to license verb phrase ellipsis, as illustrated in 32a and 32b, respectively. Ellipsis with infinitival forms of modals is marginal at best, as shown in 32c.
- (32)
a.
b.
c.
Thus, the generalization is that infinitival forms of auxiliaries do not license verb phrase ellipsis.Footnote 16 Although there may be an explanation for this restriction, for now we simply observe that verb phrase ellipsis must be licensed in the extended verbal projection by some head with inflectional features (Lobeck 1995:141–150, Merchant 2001:60), and that the class of such heads may be different in different languages. In Danish, it seems that auxiliaries in their infinitival forms do not license verb phrase ellipsis, while auxiliaries in their finite and participial forms do.
Within Platzack’s analysis of g⊘re as v, it is entirely unexpected that the infinitival form of g⊘re should prevent ellipsis. For Platzack, ellipsis versus pronominalization is mediated by the feature content of V. In his system, v bears a valued uninterpretable Infl feature that is checked by the corresponding interpretable feature on T. In Danish, V optionally also bears this uninterpretable feature. When it does not and when V also does not raise to v, then the VP is elided. If it does, and V still does not raise to v, then the VP is replaced by the proform det. Since the relationship between g⊘re and ellipsis is completely governed by a feature on V, it is not possible, within Platzack’s system, to rule out ellipsis with just the infinitival form of g⊘re. Either it should be possible with every form, or it should not be possible at all. Even supposing that an analysis of auxiliary g⊘re as v were able to derive the facts in 28, it would miss a generalization. It is not just auxiliary g⊘re that fails to license ellipsis when it appears in the infinitive: all auxiliaries display this restriction. Ideally, however we explain the lack of ellipsis with infinitival g⊘re, we would want the same analysis to extend to all the other auxiliaries. By uniting them all in a single category, our analysis does exactly this.
4. The Three Environments for Auxiliary G⊘re
We have argued that g⊘re is an auxiliary, though it is clearly different from other auxiliaries in two important respects that are related. We propose that g⊘re is a defective auxiliary since it has no dedicated semantic content and its distribution is syntactically restricted (in the same way that expletive pronouns are restricted to certain syntactic environments). In terms of its distribution, g⊘re occurs in three main environments: clauses with verb phrase ellipsis, clauses with verb phrase topicalization, and clauses where the verb phrase is replaced by the verbal proform det. It does not occur when the verb phrase is canonically realized. This contrasts with the other auxiliaries of the language that have semantic content and whose syntactic distribution is not restricted.
We propose to account for g⊘re’s defective status by restricting it to subcategorizing for only pronominal vPs, as stated in 33. All other Danish auxiliaries can subcategorize for any type of vP, including pronominal ones.
(33)
The pronominal vP that g⊘re subcategorizes for receives its interpretation in one of two ways. Either it is anaphoric and has an antecedent in the discourse, or it is bound and saturated by a sentence-initial vP. Since fully realized vP’s are not pronominal, g⊘re never co-occurs with a main verb. Thus, 34 is ungrammatical regardless of the choice of nonfinite inflection on the lower verb.
(34)
While it might strike some as odd to distinguish between pronominal and nonpronominal members of a category in this way, we know from other domains that syntax can make reference to such a distinction. For example, the Scandinavian languages are well known for object shift, an operation where the object moves to the left of the verb. In Danish, only simple pronominal DPs undergo object shift. Full DPs never do, no matter what their prosodic properties are (Vikner 1989).
Crucially, while g⊘re subcategorizes for a proform, it does not specify whether or not this proform should be overt. Nor does g⊘re sub-categorize for a single type of overt proform—det, for instance. As we show in the following sections, both overt and null proforms are attested with g⊘re, as well as several types of overt proforms.
4.1. Verb Phrase Ellipsis
Following Dalrymple et al. 1991, Hardt 1993, Lobeck 1995, and Culicover & Jackendoff 2005:266–272, we treat the elided vP as a null proform that receives its interpretation from the surrounding discourse (see Chung et al. 1995 for a related but not identical analysis of sluicing).Footnote 17 Though it has received little attention in the literature, verb phrase ellipsis is productive in Danish, as attested by the naturally occurring examples in 35.Footnote 18
- (35)
a.
b.
c.
If in ellipsis contexts, the vP is a proform, then g⊘re should be able to take it as complement, as indeed it does in 36 (see below). The second conjunct of the sentence in 36 has the structure in 37. The vP in the first conjunct functions as the antecedent of the null proform that replaces the vP, which accounts for the interpretation of the second conjunct, namely, that Mona washed the car.
One possible objection to this analysis is that the subject must merge directly in SpecTP and therefore is not interpretable. We must abandon the Uniform Theta Assignment Hypothesis (UTAH; Baker 1988) as a syntactic principle since agent arguments are no longer required to generate in Spec vP.Footnote 19 However, the basic idea behind the UTAH can be preserved. The vP proform in ellipsis contexts must be of type <e,t>—that is, a function from individuals to truth values. The e-type argument is saturated by the referent of the subject DP. If the antecedent of an anaphoric proform is an unaccusative vP, then the subject would be a patient. If the antecedent of an anaphoric proform is an unergative or transitive vP, then the subject would be an agent.Footnote 20
(36)
(37)
4.2. Verb Phrase Pronominalization
The second environment for g⊘re involves an overt verbal proform, det.Footnote 21 The proform tends to move to clause-initial position, as in 38a. However, when SpecCP is occupied by another element, such as the sentence-level adverb in 38b det occurs where canonical vPs do.
- (38)
a.
b.
We assume that det originates as the sister of an auxiliary. Thus, when SpecCP is occupied by some other element, such as the null question operator of the polar question in 39b, it does not move.
- (39)
a.
b.
(40)
When there is nothing in SpecCP, det raises, so that the sentence in 41 has the structure in 42.Footnote 22
(41)
(42)
One advantage of this analysis of verb phrase pronominalization is that it also accounts straightforwardly for another construction with g⊘re, namely verb phrase left dislocation, as shown in 43 (see Källgren & Prince 1989 for a discussion of the parallel construction in Swedish).
(43)
In addition to the fronted verbal proform det, there is a fully realized vP in a left-peripheral position. This produces verb-third word order. As shown in 44, we analyze left dislocation as verb phrase pronominalization with an overt vP adjoined to the root clause.
(44)
As a specific instance of verb phrase pronominalization, the presence of g⊘re in left dislocation contexts is licensed by the subcategorization frame in 33 since the sister of g⊘re is again a pronominal.
4.3. Verb Phrase Topicalization
Finally, much like verb phrase left dislocation, we analyze verb phrase topicalization in Danish as base-merger of a pronominal vP below an auxiliary, with the initial vP originating as an adjunct to CP. Thus, the topicalized clause in 45 has the structure in 46.
(45)
(46)
This analysis is similar to the one that Zwart (1997b) proposes for topicalization in Dutch, where a (sometimes null) pronominal element raises to SpecCP, while the topicalized phrase is adjoined to the root clause.
In principle, a simpler analysis of verb phrase topicalization—one that does not involve movement—could be considered. The initial vP is simply base-merged in SpecCP, though it is coindexed with a proform merged under an auxiliary, such as g⊘re:
(47)
However, such an analysis is not tenable because Danish verb phrase topicalization shows positive traits of movement. In this respect, verb phrase topicalization is similar to verb phrase left dislocation—an important comparison since under our analysis, the two constructions are structurally identical. For instance, with regards to island constraints the initial vP in both topicalization and left-dislocation constructions cannot be interpreted as embedded in a sentential subject, as in 48, an adjunct, as in 49, an embedded interrogative clause, as in 50, or the left or right conjunct of a coordinate structure, as in 51 and 52. Throughout, the a-examples involve topicalization, and the b-examples involve left dislocation. The underscore marks the base position of the proform.
(48) Sentential Subject Constraint
a.
b.
(49) Adjunct Island
a.
b.
(50) Wh-Island
a.
b.
(51) Coordinate Structure Constraint (extraction of left conjunct)
a.
b.
(52) Coordinate Structure Constraint (extraction of right conjunct)
a.
b.
Under our analysis, these island violations arise because a pronoun originates inside each of these islands and then raises to the matrix SpecCP, crossing an island boundary. In the a-examples, the proform is null, and in the b-examples it is det.
Again, topicalization and left dislocation behave in the same way with respect to binding conditions. In both constructions, the initial vP is interpreted where ordinary vPs are base generated, under an auxiliary.
(53) Condition A
a.
b.
(54) Condition B
a.
b.
(55) Condition C
a.
b.
As 53 shows, in both constructions, a reflexive pronoun in the sentence-initial vP can be bound by the subject DP. Moreover, as 54 shows, a nonreflexive pronoun in the same position cannot be coreferential with the subject DP. Finally, as 55 shows, an R-expression in the fronted vP, is not free when it is coreferential with the subject DP. The judgment patterns in 53–55 point to reconstruction effects. In each case, the binding possibilities are exactly as they would be if the vP occupied its canonical position below the matrix subject Peter/han. Under our analysis, the reconstruction effects cannot be due to actual reconstruction of the initial vP. Instead, these effects fall out from movement of the proform.
Two further movement diagnostics—the licensing of parasitic gaps and crossover effects—turn out to be inconclusive. The grammaticality of 56, with or without an overt proform, suggests that both fronting constructions license parasitic gaps, and hence involve movement.
(56)
However, since the gap corresponds to a vP and the initial vP provides an antecedent, it could simply be an instance of verb phrase ellipsis. Support for this claim comes from the fact that the second gap is not dependent on the first, as the grammaticality of 57 makes clear.
(57)
Unless the possibility of ellipsis in the second clause can be eliminated, the grammaticality of examples such as 56 does not tell us anything about the derivation of topicalization and left dislocation in main clauses.
It is difficult to establish crossover effects because strong crossover requires that the element being crossed over c-command the base position of the crossing element. In the case of topicalization, this means that a configuration must be constructed in which one vP proform det c-commands the base position of another vP proform (which can be null or overt). The only such configuration that we have been able to identify is coordination of a vP with another vP containing a vP proform. Topicalization out of that second conjunct is indeed ungrammatical, but it is already ruled out by the Coordinate Structure Constraint in 51 and 52. Thus, it is impossible to tell whether verb phrase topicalization yields strong crossover effects.
Weak crossover configurations (where the crossed-over element does not c-command the base position of the crossing element) can be constructed, as in 58. In this example, the relative clause modifying the subject contains a vP proform det. This proform is coindexed with the proform, which has the initial vP as its antecedent. If the latter moves from its original position below negation in the main clause to a position before the finite auxiliary it would cross over a coindexed proform, namely det in the relative clause.
(58)
Speakers find both versions of 58 with and without the proform repetitive and somewhat strained, but grammatical. On the one hand, the lack of a crossover effect in 58 could be interpreted as evidence against a movement analysis of verb phrase topicalization. On the other hand, weak crossover effects are notoriously variable and hard to identify even with nominal proforms, as noted by Wasow (1979:157–175), which suggests that it is not a very reliable diagnostic. The island and connectivity effects documented above all support the movement analysis, and this evidence seems to be conclusive enough, so that the data in 58 could be set aside.
4.4. Extensions
Our proposal predicts that g⊘re should be possible when the vP is replaced by any proform, not just the ones we have considered. This is indeed the case. In 59, the vP is replaced by the relative pronouns hvad ‘what’ and hvilket ‘which’—which raise to SpecCP—and g⊘re is acceptable in these examples.Footnote 23
- (59)
a.
b.
In 59a, g⊘re is clearly not the main verb since the relative clause modifies a stative predicate, holde kaje ‘keep quiet’, and g⊘re does not modify this predicate’s thematic roles (see criterion iv in section 2.2).
Our proposal finds counterparts in other frameworks. Working within Lexical Functional Grammar, L⊘drup (1990) treats gj⊘re in Norwegian as an auxiliary that, just like its Danish counterpart, appears when the verb phrase is topicalized or pronominalized, as shown in 60 and 61.
(60)
- (61)
a.
b.
To capture its distribution, L⊘drup (1990:10) imposes a restriction on gj⊘re that he refers to as R: “the verbal complement of gj⊘re must enter into an unbounded dependency.” Restriction R accounts for the presence of gj⊘re in 60 and 61b, where the (pronominalized) vP has indeed been fronted. However, in order to account for sentences such as 61a, where det stays in situ, L⊘drup proposes that the proform is exempt from R. Though he does not consider such examples, R would not derive the occurrence of g⊘re in sentences where the vP has been elided, which is possible in Norwegian as illustrated in 62.
(62)
In contrast, our treatment of g⊘re as a defective auxiliary—that is, an auxiliary that only subcategorizes for pronominal vPs—successfully unifies all of these environments, as well as extends to other syntactic environments where the vP is replaced with an overt proform.
5. Conclusion
The major analytical challenges presented by Danish auxiliary g⊘re are its limited but regular distribution and its position relative to adverbs and other auxiliaries. We have argued that these challenges are met by an analysis that treats g⊘re as a defective auxiliary. As an auxiliary, g⊘re must find its place in the extended verbal projection, and given the distribution of nonfinite forms of g⊘re, we concluded that it is located at the very bottom of the auxiliary hierarchy. This low position makes it difficult to distinguish our auxiliary analysis from Platzack’s analysis of g⊘re as the realization of v. A wider range of data—that is, the distribution of g⊘re with other auxiliaries—supports our analysis of g⊘re as a member of the category Aux. The fact that g⊘re cannot follow perfect være or passive blive suggests that either være and blive are not auxiliaries (a highly suspect conclusion), or that g⊘re is not the realization of v (as we argue). In addition, the infinitival form of auxiliary g⊘re does not license vP ellipsis. Again, this remains unexplained if g⊘re is v. If, instead, it is an auxiliary it falls in line with other auxiliaries, which exhibit the same restriction.
A common intuition about English do-support is that it is, as Grimshaw (1997:381) puts it, “possible only when it is necessary.” Within a transformational framework, do-support is a kind of last resort strategy—it is an operation that applies at the end of a failed derivation to save it.Footnote 25 Although our analysis rejects a similar characterization of auxiliary g⊘re in Danish it is worth asking what such a last-resort analysis would look like. First of all, auxiliary g⊘re would never be present in the narrow syntactic representation of a sentence. It would instead be inserted to host some inflectional material whose normal host is unavailable because of the details of a particular derivation. For example, g⊘re would appear when the verb has been manipulated through ellipsis, pronominalization, or topicalization. By contrast, in our analysis, the restricted distribution of auxiliary g⊘re follows from its defective status. Auxiliary g⊘re is part of the Danish lexicon—it is a feature bundle that can be added to the numeration, just like any other auxiliary, and it can be merged into the extended verbal projection, again, just like any other auxiliary. Unlike other auxiliaries, however, it only takes a pronominal vP complement.
Do these analyses of auxiliary g⊘re yield any empirical differences? If we consider finite g⊘re alone, the answer seems to be no. Tense must be expressed, and, in the relevant contexts, it has to be expressed on g⊘re because no other verbal form is present to host the tense suffix. Whether g⊘re is present in the structure from the beginning of the derivation (as in our base-generation analysis) or recruited at a relatively late stage in the derivation (as in the last-resort analysis) appears to be a matter of analytical preference and theoretical commitment. However, the existence of nonfinite g⊘re, which is generally optional, poses a challenge for last-resort analyses. If nonfinite g⊘re is optional, as it is in many contexts, then nonfinite inflectional morphology does not necessarily have to be expressed overtly. If it does not have to be expressed, then it is not clear what would motivate the insertion of nonfinite g⊘re. If there is no problem to solve, then there is no need to resort to any strategy, let alone a last-resort one. Since most dialects of English lack nonfinite forms of auxiliary do, this issue has not figured prominently in the literature (though, see Baltin 2007 on British English). In contrast, the pervasiveness of nonfinite g⊘re in Danish brings this issue to the forefront and suggests that a base-generation account is on the right track, at least for Danish and related languages.
