2. OS OF VERBAL ANAPHORS

Anaphors differ in the semantic type of their antecedent. The antecedent of the anaphor in (1b) above is entity-denoting, type <e>, while the antecedents in (7) and (8) are propositional (marked with square brackets in the English context sentence), type <t> and <e,t>, respectively.

1. (7)

   

2. (8)

   

Andréasson (Reference Andréasson, Butt and King2008, Reference Andréasson, Butt and King2010) shows that propositional anaphors illustrated in (7) and (8) are more reluctant to shift than entity-denoting anaphors (even though the anaphor in (8) is actually shifted). They can occur in situ without receiving a contrastive interpretation, as in (9a), and they can even be marginal in the shifted position, as in (9b). An entity-denoting anaphor on the other hand canonically receives a contrastive interpretation in situ, see (10a).

1. (9)

   

2. (10)

   

Andréasson relates this difference in interpretation of in-situ anaphors to cognitive accessibility. On her analysis the shifted position is associated with a discourse participant in focus, while the in-situ position is associated with an activated discourse participant.Footnote ³ In order to be in focus (and thus in the shifted position), the participant must be easily accessible. The anaphor det ‘it’ in (9) is less accessible since it has a propositional antecedent and since it occurs in a non-assertive environment (a polar question). Due to decreased accessibility, the anaphor is in situ without being contrastive. The anaphor ham ‘him’ in (10) is cognitively more easily accessible having an entity-denoting antecedent and it receives a contrastive interpretation in situ.

Andréasson supports this analysis by showing that propositional anaphors are often in situ without being contrastive in a wide range of non-assertive environments: polar questions (11), imperatives (12), conditional clauses (13) (all V1 clauses) and finally wh-questions (14).

1. (11)

   

2. (12)

   

3. (13)

   

4. (14)

   

Andréasson (Reference Andréasson, Butt and King2010) also observes that propositional anaphors are more reluctant to be in the shifted position with non-factive verbs than with factive verbs. Again this is explained with reference to cognitive accessibility. A complement presupposed to be true, must be cognitively easily accessible.Footnote ⁴ In conjunction with non-factive verbs, propositional anaphors only shift if another element of the clause has contrastive focus (see Andréasson Reference Andréasson2013).

Andréasson's investigation raises some new and challenging questions about OS in Danish and it also faces some problems. Andréasson (Reference Andréasson, Butt and King2008) sets out to discuss anaphors with propositional antecedents, but she ends up concentrating on anaphors with sentential (CP) antecedents as also acknowledged in Andréasson (Reference Andréasson, Butt and King2010:29, fn. 1). Thus, her investigations do not show that VP anaphors are subject to the same constraints as anaphors with CP antecedents. In fact, I will show that VP anaphors do not behave as predicted by the analysis in Andréasson (Reference Andréasson, Butt and King2008, Reference Andréasson, Butt and King2010), but my focus on VP anaphors is only motivated by space limitations. I suspect that this account also carries over to anaphors with CP antecedents, but it is a matter of future research.

Two things suggest that cognitive accessibility is not the determining factor for OS of VP anaphors or not the sole determining factor at least. The first point is that in-situ placement of propositional anaphors is not restricted to non-assertive clauses such as interrogatives and directives. In-situ placement is also found in assertive clauses, namely declarative clauses with a sentential adverb in SpecCP (Ørsnes Reference Ørsnes, Bonami and Hofherr2011:424; Mikkelsen Reference Mikkelsen2012:25), as in (15) below. Thus, the right generalization is not that these clauses are non-assertive, but rather that they have an operator in SpecCP.Footnote ⁵ This seems also to be the intuition behind the formulation in Andréasson (Reference Andréasson, Butt and King2010:38), where she speaks of environments where SpecCP is not available for the anaphor.

1. (15)

   

Footnote ⁶

The second point is that non-factive verbs with VP complements do not behave alike with respect to the placement of anaphoric objects. If object placement were determined by cognitive accessibility, we should expect all non-factive verbs with VP complements to have the anaphor in situ when SpecCP is blocked. But this is not what we find. The verbs forsøge ‘to try’ and agte ‘to intend’ both select VP complements and they allow a VP anaphor to be in situ in a non-assertive clause, as in (16a) and (17a). But the verb forsøge ‘to try’ also allows for the anaphor to shift, as in (16b), while the verb agte ‘to intend’ does not, see (17b).Footnote ⁷

1. (16)

   

2. (17)

   

Footnote ⁸Footnote ⁹Footnote ¹⁰ Similarly, the verb forsøge ‘to try’ allows a VP anaphor to shift in a declarative subject-initial clause without the subject being contrastive (contrary to the analysis in Andréasson Reference Andréasson, Butt and King2010). In (18) the subject of forsøgte ‘tried’ is elided.

1. (18)

   

Footnote ¹¹ The verb agte ‘to intend’ does not allow post-verbal VP anaphors in subject-initial clauses at all, no matter whether the subject is contrastive or not.

1. (19)

   

Footnote ¹² In order to maintain cognitive accessibility as the determining factor, we would have to assume that the complement of forsøge ‘to try’ is intrinsically in focus (in the terminology of Andréasson), while the complement of agte (and many other verbs with VP complements) is intrinsically activated, and thus in situ. But still, this would not explain why the complement of agte ‘to intend’ cannot be in situ in subject-initial declaratives as in (19b) above. Thus, the questions that the analysis in Andréasson (Reference Andréasson, Butt and King2008, Reference Andréasson, Butt and King2010) fails to answer for Danish are:

• • Why do some verbs allow a VP anaphor to shift in clauses with a (possibly silent) operator in SpecCP, while other verbs do not?

• • Why do some verbs allow post-verbal VP anaphors in subject-initial V2 clauses, while other verbs do not?

I will suggest that it is a matter of subcategorization and I will prove this by first looking at the distribution of non-finite do-support in Danish. Non-finite do-support is illuminating, because the exceptions to otherwise obligatory non-finite do-support are exactly the syntactic environments identified by Andréasson as allowing a VP anaphor to be in situ without a contrastive interpretation. Non-finite do-support is obligatory with verbs that do not allow post-verbal anaphors in subject-initial clauses, but it is optional with the very same verbs in case SpecCP is occupied by an operator. Thus, it is to be expected that an analysis of non-finite do-support can help to understand why some anaphors are not interpreted contrastively in situ.

3. OS AND NON-FINITE DO-SUPPORT

Ørsnes (Reference Ørsnes, Bonami and Hofherr2011) – among other things – investigates the use of non-finite gøre ‘to do’ with verbal anaphors as illustrated in (20).Footnote ¹³

1. (20)

   

In (20) the VP anaphor det ‘it’ is the complement of the infinitival dummy verb gøre ‘to do’ and not of the modal ville ‘will’. This is referred to as non-finite do-support. Non-finite do-support is observed with all verbs selecting VP complements: auxiliaries, modals, raising verbs (subject or object) and control verbs.Footnote ¹⁴ These verbs fall into different groups as to whether they allow an NP complement to alternate with the VP complement. The first group of verbs does not allow an NP complement. These are e.g. the modals skulle ‘shall.inf’, måtte ‘to be allowed to’, ville ‘will.inf’ and burde ‘ought.to.inf’, as well as the raising verbs pleje ‘to use to’, as in (21) and lade ‘to let’, as in (22).Footnote ^15, Footnote ¹⁶

1. (21)

   

2. (22)

   

The second group of verbs allows an NP complement to alternate with the VP complement. These are e.g. the modal kunne ‘can’, as in (23), and the control verbs savne ‘to miss’ and forsøge ‘to try’, the latter illustrated in (24).

1. (23)

   

2. (24)

   

The third group of verbs also allows NP complements, but only on a special use or a special reading of the verb. These are e.g. the verbs være ‘to be’, have ‘to have’ and blive ‘to become/to get’. They only allow VPs in their use as auxiliaries, but also NPs in their use as main verbs or copula verbs. The control verbs agte ‘to intend/to honour’ and tillade ‘to allow’ also belong to this group. The verb agte only allows a VP on the reading ‘to intend’, but an NP on the reading ‘to honour’. The verb tillade ‘to allow’ only allows an NP complement when used as a strict transitive verb, as in (25), but only a VP complement when used as an object control verb, as in (26).

1. (25)

   

2. (26)

   

For ease of exposition I will concentrate on subject-raising verbs and subject-control verbs in the following discussion.

The example in (19) shows that the subject-control verb agte ‘to intend’ does not allow an anaphor as a complement in a subject-initial clause. With non-finite do-support the example is impeccable.Footnote ¹⁷

1. (27)

   

If, however, the verbal anaphor is in SpecCP, non-finite do-support is only optional (but preferred in environments such as the counterfactual clause in (20), see Ørsnes Reference Ørsnes, Bonami and Hofherr2011:422ff.).

1. (28)

   

Ørsnes (Reference Ørsnes, Bonami and Hofherr2011) proposes the generalization in (29) for non-finite do-support in Danish.

1. (29)

   

As indicated, this generalization is explained as a constraint on subcategorization. The verb agte ‘to intend’ only allows a VP as its complement and this subcategorization requirement is violated in (19), where an NP det ‘it’ occurs in complement position (either shifted or in situ).Footnote ¹⁸ In (27) the dummy verb gøre ‘to do’ takes the VP anaphor as its complement and projects a VP thereby satisfying the subcategorization requirements of the verb.Footnote ¹⁹

The question is then why this verb allows an NP complement if the complement is in SpecCP, as in (28). Ørsnes (Reference Ørsnes, Bonami and Hofherr2011) suggests that this is an instance of a movement paradox. In a movement paradox a filler does not match the syntactic category of its gap (Bresnan Reference Bresnan2001, Webelhuth Reference Webelhuth and Müller2007, Kim Reference Kim2011, a.o.). Bresnan (Reference Bresnan2001:17) gives the following example where the filler is a CP and the gap is an NP (the object of the preposition):

1. (30)

   

In (31) the filler is an NP, while the gap is a VP.

1. (31)

   

The generalization appears to be that constituents in SpecCP are allowed to relax subcategorization requirements by being of a different category than required by the verb, while constituents in the canonical complement position after the verb must match the selection of the verb. Since SpecCP is canonically associated with special discourse prominence, we can state this in broader terms: A discourse-prominent constituent is allowed to be a non-canonical complement, i.e. a complement that does not match the subcategorization requirements of the verb.Footnote ²⁰ This explains the contrast between (32a) on the one hand and (32b) and (32c) on the other hand.

1. (32)

   

But why, then, are VP anaphors nevertheless allowed to occur post-verbally in sentences like (33) with an operator in SpecCP?

1. (33)

   

Ørsnes (Reference Ørsnes, Bonami and Hofherr2011) proposes that this is due to the fact that the anaphor is prevented from being in SpecCP (this idea is also inherent to the account in Andréasson Reference Andréasson, Butt and King2010, but it is not spelled out in detail). Consider the environments where non-finite do-support is only optional (and where the VP anaphor occurs in situ, as noted by Andréasson Reference Andréasson, Butt and King2008).

1. (34)

   

In these environments the anaphor is prevented from being in SpecCP for independent reasons: In the V1 clauses in (34a–c) there is no first position (or it is occupied by a silent operator), and in (34d) and (34e) SpecCP is filled by an overt operator. Thus, these verbs license post-verbal anaphors when the anaphor is prevented from being in the first position for independent reasons. The anaphor should be in SpecCP, but stays in the post-verbal domain in order for an operator to be in SpecCP. Thus, non-finite do-support is optional when a non-canonical complement cannot be in SpecCP because an operator has to be there. Still this does not answer the question why the anaphor is in situ and not shifted. A first step towards an explanation is to take a close look at the contrast in (35) and (36).

1. (35)

   

2. (36)

   

As (35) shows, the verb savne ‘to miss’ allows a VP anaphor to shift and to be in situ. The verb pleje ‘to use to’ on the other hand must have the anaphor in situ. Moreover, the verb savne ‘to miss’ does not require do-support in a subject-initial clause, while pleje ‘to use to’ does; consider (37) and (38).Footnote ²¹

1. (37)

   

2. (38)

   

The contrast between (37) and (38) is explained on the analysis in Ørsnes (Reference Ørsnes, Bonami and Hofherr2011): The verb savne ‘to miss’ allows its VP complement to alternate with an NP complement as shown in (39), while pleje ‘to use to’ does not, see (40). Therefore savne ‘to miss’ can take a pronoun as its complement, while pleje ‘to use to’ cannot.

1. (39)

   

2. (40)

   

This tells us something important about OS in Danish: OS is only possible if the verb allows an NP complement (or a locative, see below). The verb in (37) allows an NP object, while the verb in (38) does not.Footnote ²² It follows that OS can indeed be disambiguating in the cases where a verb allows an NP complement on a special reading. If the verb occurs with OS, it will be associated with the reading of the verb where it allows an NP complement. In (41) the preferred reading of the verb agte is ‘to honour’.

1. (41)

   

Returning to Andréasson's analysis: What on the analysis in Andréasson (Reference Andréasson, Butt and King2008, Reference Andréasson, Butt and King2010) are environments with anaphors in situ due to cognitive (in)accessibility, are environments where a pronominal not matching the subcategorization of the verb is allowed to be in complement position because no other position is available. SpecCP is blocked and the pronominal cannot shift since a shifted pronoun must match the subcategorization requirements of the verb.

The claim that shifted pronouns must be subcategorized is challenged by shifting locatives as in (42) (see also Hansen & Lund Reference Hansen and Lund1983:42).

1. (42)

   

Engels & Vikner (Reference Engels and Vikner2012:85) and Mikkelsen (Reference Mikkelsen2011:249) analyze shifted locatives as adjuncts, but there is evidence that also shifted locatives are subcategorized by the verb. In KorpusDK shifted locatives are primarily found with the following verbs: arbejde ‘to work’, blive ‘to stay’, bo ‘to live’, komme ‘to come’ and være ‘to be’. All these verbs can be argued to subcategorize a locative (see also Hansen & Heltoft Reference Hansen and Heltoft2011:1017f.). One piece of evidence is that subcategorized locative PPs allow the complement of the preposition to be extracted, while extraction out of an adjunct locative is degraded, as shown in (43).

1. (43)

   

The verbs found with shifting locatives allow preposition stranding with a locative PP.

1. (44)

   

Moreover, the verbs arbejde ‘to work’ and læse ‘to read’ only select a locative on a special reading, namely ‘to be employed at’ and ‘to be enrolled at’. Thus, (44a) can only mean that the speaker is employed at the institution in question.Footnote ²³ Consider also the contrast between the two uses of læse ‘to read’ in (45). Preposition stranding is impeccable on the reading ‘to be enrolled at’, as in (45a). On the other hand, (45b) is odd, since one cannot be enrolled at a library.

1. (45)

   

With a shifted locative, arbejde ‘to work’ can only mean ‘to be employed at’, as in (42), and læse ‘to read’ can only mean ‘to be enrolled at’, as in (46).Footnote ²⁴ This is expected if a shifted pronoun must be subcategorized by the verb.Footnote ²⁵

1. (46)

   

Footnote ²⁶ So far we have established that in order for a pronoun to undergo OS, it must be subcategorized by the verb (see however note 24 on ethical accusatives) and it must match the category selection by the verb. Only verbs subcategorizing NPs and (locative) ADVPs allow OS. If a verb only subcategorizes a VP and this VP is realized as the VP anaphor det ‘it’, the anaphor must be in SpecCP or in situ in case SpecCP is occupied by an operator. In the next section I will discuss the impact of the information structural properties of VP anaphors on their linearization with verbs that allow shifting and verbs that do not.

4. THE INFORMATION STRUCTURE OF THE OBJECT POSITIONS

The preceding discussion raises an important question: Why must a shifted pronoun meet the category selection of the verb? As a possible explanation I will examine the information structural properties associated with the three possible object positions: SpecCP, shifted and in situ.

I follow Mikkelsen (Reference Mikkelsen2011, Reference Mikkelsen2012 among others) in assuming that SpecCP and in situ are discourse-prominent positions. SpecCP is canonically the position of topics (discourse-old, salient information), while in situ is the position of a contrastive focus (presupposing a salient set of alternatives). The shifted position in turn is associated with discourse-old, non-salient information, hence backgrounded information.Footnote ²⁷

The information structural properties of the various object positions can be illustrated with presuppositional verbs. The verb savne ‘to miss’ carries two presuppositions (PRs): (i) The subject referent has engaged in the activity (in a broad sense) denoted by the VP before, and (ii) the subject referent is not engaged in the activity denoted by the VP at present. This is illustrated in (47).

1. (47)

   

In the following I will concentrate on the negative presupposition, called PR₁ in (47). If the verb savne ‘to miss’ occurs with a VP anaphor, there is an interesting difference in the shifting behaviour of the anaphor according to whether the negative presupposition has been introduced in the discourse already or not. Consider first the example in (48).

1. (48)

   

Footnote ²⁸ The antecedent for det ‘it’ is ‘to sing in the evening’. The presupposition that the speaker does not sing in the evening (at present) is not stated explicitly, it has to be accommodated by the hearer (Lewis Reference Lewis1979, Stalnaker Reference Stalnaker2002, von Fintel Reference von Fintel2008). In that case the continuation in a. (the one found in the original example), with the anaphor in SpecCP, is much better than the continuation in b., with the shifted anaphor. Consider next example (49).

1. (49)

   

In (49) the antecedent of the VP anaphor is ‘to drink coffee’. The presupposition that the speaker no longer drinks coffee is explicitly introduced into the discourse by the verb ‘to stop’ (in Danish at holde op) so no presupposition accommodation (PA) is required. In this case the continuation in a., with the anaphor in SpecCP, is odd, while the continuation in b. (the one found in the original example), with the anaphor in the shifted position, is impeccable. In (50) the fact that the speaker does not engage in having vacation is also introduced explicitly in the discourse and again we see that the VP anaphor det ‘it’ with the antecedent ‘to have vacation’ shifts.

1. (50)

   

It seems from these examples that SpecCP is preferred when the VP anaphor is associated with PA, while the shifted position is preferred when no PA is needed. PA is a hearer-initiated pragmatic operation so it places an increased processing burden on the hearer (Schwenter & Waltereit Reference Schwenter, Waltereit, Davidse, Cuyckens and Vandelanotte2010). This means that if the complement of savne ‘to miss’ is an anaphor as in the examples above, the hearer not only has to identify the antecedent of the VP anaphor, the hearer also has to accommodate a negative presupposition, i.e. to infer the opposite polarity of the antecedent as the relevant presupposition. It is thus to be expected that accommodation of the negative presupposition of savne ‘to miss’ must pertain to salient information.Footnote ²⁹ From a processing perspective it is counter-intuitive to background information and – at the same time – to invite the hearer to invest increased processing efforts in inferring new information about this backgrounded information. Consequently PA should be facilitated when it pertains to information in a discourse-prominent position. The fact that PA is facilitated when a VP anaphor is in SpecCP, while it is degraded when the anaphor is shifted, shows that SpecCP is discourse-prominent, while the shifted position is not. Note that these data are not explained by the analysis in Mikkelsen (Reference Mikkelsen2012), where the VP anaphor is claimed to be inherently topical and to be in SpecCP when the subject is information-structurally indistinguished and to be in the post-verbal domain when the subject is information structurally salient. The subjects in (48) and (49) have the same information structural status.

Next we turn to the in-situ position containing contrastive information. Presupposition accommodation can also target a VP anaphor in situ.

1. (51)

   

In (51) the antecedent of the VP anaphor is ‘to sing in a choir’. The presupposition that the speaker no longer sings in a choir is not explicitly given. It is accommodated. In situ is favoured because the antecedent ‘to sing in a choir’ is contrasted with the VP ‘to sing’, but not ‘in a choir’. The fact that PA can target an anaphor in situ as in a. confirms that in situ contains salient information. To sum up: PA is felicitous with anaphors in SpecCP and in situ, but not in the shifted position. Thus, the shifted position contains non-salient, while the other two positions contain salient, information.

Now if OS is associated with backgrounded (hence non-salient) information and in situ is associated with contrastive (salient) information, how shall we interpret the information structural status of a VP anaphor in situ, if no other post-verbal position is available? Is the anaphor in situ necessarily contrastive, or does it mean that these verbs allow a non-contrastive anaphor to be in situ, as claimed by Andréasson (Reference Andréasson, Butt and King2008:33–34).

Consider the two verbs savne ‘to miss’ and pleje ‘to use to’ in a polar question where SpecCP is unavailable.

1. (52)

   

In the interrogative response of B in (52), the VP anaphor det ‘it’ represents old, backgrounded information and it is shifted. Note also that no PA is required since the negative presupposition is given. In (52b) in situ is odd, since it invites a contrastive reading in a context with no salient set of alternatives. Consider next the very same discourse with the verb pleje ‘to use to’.

1. (53)

   

The discourse in (53) presents the very same information structural setting as (52), but here the VP anaphor has to be in situ, even though there is no basis for a contrastive reading and no accommodation of a negative presupposition is required. The anaphor contains old, non-salient information, but the verb pleje ‘to use to’ does not subcategorize an NP complement so the anaphor is not allowed to shift. The anaphor is in situ, where it receives stress (Hansen Reference Hansen1977:60), but it does not receive a contrastive interpretation, as also noted in Andréasson (Reference Andréasson, Butt and King2008:33–34).

Consider finally a contrastive context for these two verbs. In (54) a contrast is established between ‘to smoke’ and ‘to drink wine’ and the question is which of the two Peter is missing. As expected, in situ is appropriate for the anaphor in connection with the verb savne ‘to miss’.

1. (54)

   

In a contrastive setting, in situ is also perfectly acceptable for the anaphor with pleje ‘to use to’.

1. (55)

   

Thus, the conclusion is that if a non-canonical complement is forced to be in the post-verbal complement position, it must be in situ where it canonically receives stress. The information structural status of the anaphor as either backgrounded or contrastive does not matter. Thus, for non-canonical complements the information structural distinction between shifted and in situ is suspended. The conclusion is that a non-canonical complement must be syntactically prominent in Danish – either by being in SpecCP or by being in situ, where it receives stress. Just as backgrounded information cannot be the target of the accommodation of a negative presupposition, backgrounded information cannot be associated with a non-canonical complement. This notion of discourse prominence for non-canonical complements bears some resemblance to the notion of cognitive accessibility in Andréasson (Reference Andréasson, Butt and King2008, Reference Andréasson, Butt and King2010). Still, while Andréasson argues that propositional anaphors tend to be discourse-prominent (in situ) because they have propositional – and hence cognitively less accessible – antecedents, I argue for Danish that the kind of antecedent is not the sole determining factor. I argue that some VP anaphors must be discourse-prominent (in SpecCP or in situ under special circumstances) because they are non-canonical, i.e. when they are not categorially selected by the verb.Footnote ³⁰ This explains that VP anaphors can shift with some verbs, but not with others.

5. TOWARDS AN OT ANALYSIS OF THE PLACEMENT OF VP ANAPHORS

Optimality Theory (OT) assumes that a grammatical sentence is the result of evaluating a set of possible candidates for a certain input against a set of ranked constraints (Prince & Smolensky Reference Prince and Smolensky1993). The optimal candidate – and thus the grammatical output of the evaluation – is the candidate incurring the fewest costly violations. For a sentence to be grammatical does not mean that it does not violate any constraints of the language. It means that it violates fewer or less expensive constraints than other possible candidates for a certain input. Several features of the analysis so far suggest an account in terms of violable constraints: In Danish a VP anaphor can remain in the post-verbal domain, only if SpecCP is occupied by an operator and a VP anaphor can be in situ without receiving a contrastive interpretation, but only if it is a non-canonical complement, that is if it does not match the categorial selection by the verb.

The analysis so far has uncovered that only subcategorized pronouns meeting the categorial selection of the verb are allowed to shift. However, a prerequisite is of course that Holmberg's Generalization holds. Engels & Vikner (Reference Engels and Vikner2012, Reference Engels and Vikner2013) propose an OT analysis of Holmberg's Generalization as a (violable) constraint on order preservation. They assume a constraint to the effect that a bare, non-focal pronoun must shift out of the focal VP domain. However, OS can lead to a violation of a higher-ranked constraint stating that an object must follow its selecting head (in Danish), as in (56a), and that an indirect object must precede a direct object, as in (56b). This explains the ungrammaticality of the examples in (56) (Engels & Vikner Reference Engels and Vikner2012:50).

1. (56)

   

I believe that the OT analysis of Object Shift in Engels & Vikner (Reference Engels and Vikner2012, Reference Engels and Vikner2013) is compatible with the following analysis, but for ease of exposition I will omit this part from the OT tableaux. Thus, I tacitly assume that Holmberg's Generalization holds in the first place. I do, however, deviate from the analysis in Engels & Vikner in assuming that object placement is a competition between three positions: SpecCP, the shifted position and the in-situ position – as suggested in Andréasson (Reference Andréasson, Butt and King2010). The three positions (canonically) correlate with the information structure (IS) notions: topic, background and contrast, respectively. Since a VP anaphor can have either of these information structural values, we have to consider three different inputs for each verb: one with a topical, one with a backgrounded and one with a contrastive anaphor.

The sample verbs will be the subject-raising verb pleje ‘to use to’, which does not allow an NP complement, and the subject-control verb savne ‘to miss’, which does allow an NP complement.

1. (57)

   

2. (58)

   

Moreover we have to consider two environments for each of these verbs according to whether SpecCP contains an operator or not. The first environment will be a non-operator clause, namely an assertive clause as in (59).

1. (59)

   

The second environment is an operator clause, namely a wh-question as in (60).

1. (60)

   

The analysis has to account for the following two generalizations about Danish uncovered in the discussion above:

1. (i) If a verb does not subcategorize an NP complement, a VP anaphor has to be in SpecCP in a non-operator clause. In an operator clause (where SpecCP is occupied by an operator) the VP anaphor has to be in situ (in the second most prominent position). The information structural value of the VP anaphor does not matter.

2. (ii) If a verb does subcategorize an NP complement, the linearization of a VP anaphor in a non-operator clause is dependent on the information structural value of the anaphor as either topical (SpecCP), backgrounded (shifted) or contrastive (in situ). In an operator clause, where SpecCP is not available, a topical anaphor is in situ (the second most prominent position). Otherwise the VP anaphor is linearized according to its information structural value: A backgrounded VP anaphor is shifted and a contrastive VP anaphor is in situ.

Thus, there is an interesting overall generalization: Whenever a VP anaphor is prevented from being in SpecCP by an operator, it has to be in situ. In a sense it has to be in the second most prominent position. Thus, there appears to be a ban on backgrounding of topical anaphors and of anaphors as non-canonical complements (complements not categorially selected by the verb).

5.1 Non-operator clauses

For non-operator clauses, where SpecCP is available for the VP anaphor, we will need the two constraints presented in Table 1.

Table 1. Constraints for OT analysis of VP anaphors in non-operator clauses.

For the verb pleje ‘to use to’, which does not allow an NP complement, we get the results presented in Tableaux 1a–c. With the verb pleje ‘to use to’, the VP anaphor must be in SpecCP regardless of the information structural value. This is the only way to avoid violation of the high-ranked constraint that a post-verbal complement must be canonical. In two cases the price is a violation of the lower-ranked constraint on a canonical association of syntactic positions with information structural values since SpecCP is not a topic.

Tableau 1a. Placement of topical VP anaphor in non-operator clause with VP-selecting verb.

Tableau 1b. Placement of backgrounded VP anaphor in non-operator clause with VP-selecting verb.

Tableau 1c. Placement of contrastive VP anaphor in non-operator clause with VP-selecting verb.

For the verb savne ‘to miss’, we get the results in Tableaux 2a–c below. With the verb savne ‘to miss’ the constraint that a post-verbal complement must be canonical is never violated, since savne ‘to miss’ allows an NP complement (alternating with a VP complement). Here only the information structural value of the VP anaphor is relevant for the placement of the pronoun: A topical pronoun is in SpecCP, a backgrounded anaphor shifts and a contrastive anaphor is in situ.

Tableau 2a. Placement of topical VP-anaphor in non-operator clause with VP/NP-selecting verb.

Note: In these examples two violations are incurred: (i) SpecCP is not associated with topic, and (ii) the shifted position is not associated with backgrounding or the in-situ position is not associated with contrastive information.

Tableau 2b. Placement of backgrounded VP anaphor in non-operator clause with VP/NP-selecting verb.

Tableau 2c. Placement of contrastive VP anaphor in non-operator clause with VP/NP-selecting verb.

5.2 Operator clauses

Next we turn to clauses with an operator in SpecCP. Here two further constraints are operative, presented in Table 2.

Table 2. Additional constraints for OT analysis of VP anaphors in operator clauses.

Consider first, in Tableaux 3a–c, a wh-question with the verb pleje ‘to use to’ that only allows a VP complement (I assume that a wh-constituent is a focus). For the verb pleje ‘to use to’, the high-ranked constraint that an operator must be in SpecCP forces the VP anaphor to be in the post-verbal domain thereby incurring a violation of the constraint that a complement must be canonical to be in complement position. Within the post-verbal domain the anaphor can be shifted or in situ. Since a non-canonical complement cannot be backgrounded, candidates with a shifted VP anaphor incur a violation of the high-ranked constraint *BCK-TN. Candidates with the VP anaphor in situ do not incur this violation. Thus, a post-verbal anaphor is in situ regardless of its information structural value.

Tableau 3a. Placement of topical VP anaphor in operator clause with VP-selecting verb.

Tableau 3b. Placement of backgrounded VP anaphor in operator clause with VP-selecting verb.

Tableau 3c. Placement of contrastive VP anaphor in operator clause with VP-selecting verb.

Consider next, in Tableaux 4a–c, the verb savne ‘to miss’ in the very same environments. This verb allows an NP complement. For the verb savne ‘to miss’, the constraint VP-CAN does not take effect since the VP anaphor is a canonical complement. The constraint OP-SPC, however, takes effect to prevent a topical VP anaphor from being in SpecCP. A topical VP anaphor goes into the in-situ position to avoid backgrounding of a topic. Thereby it only violates the lower-ranked constraint RSP-IS since a pronoun in situ ought to be contrastive in a context where Object Shift is possible.

Tableau 4a. Placement of topical VP anaphor in operator clause with VP/NP-selecting verb.

Tableau 4b. Placement of backgrounded VP anaphor in operator clause with VP/NP-selecting verb.

Tableau 4c. Placement of contrastive VP anaphor in operator clause with VP/NP-selecting verb.