3.1 Contrast on the object

The primary function of contrastive focus is the evoking of alternatives. The focusing of a constituent raises the assumption of the existence of a set of alternative elements to the one expressed. This alternate set may be overt in the context or presupposed (see Rooth Reference Rooth1992). Whenever a Danish or a Swedish pronominal object has a contrast interpretation and – in speech – contrastive stress, this prevents it from shifting; instead it is licensed in the in situ position, see the Danish example in (10).

1. (10)

   

In example (10) both the contrastive stress on the object pronoun (here and throughout this paper marked with capitals) and the position of the pronoun to the right of the negation are linguistic markers of contrast. Nevertheless, the object referent is still just as accessible as it would have been in a non-contrastive context, and its actvn value is still 0.

This restriction on pronominal objects to remain in situ when contrasted is what gave rise to the analyses of object shift suggesting that an unstressed pronoun escapes from a focus domain (see Holmberg Reference Holmberg1999:23). Interestingly, when an Icelandic object pronoun has a contrastive interpretation, it does not have to remain in situ. Contrasted object pronouns are also allowed in the shifted position, see the examples in (11) and (12) from Thráinsson (Reference Thráinsson2007:32, 67).

1. (11)

   

2. (12)

   

Definite full NP objects in Icelandic are, however, constrained to remain in situ when contrasted. Collins & Thráinsson (Reference Collins and Thráinsson1996:406) state: ‘A stressed (and therefore focused) definite NP behaves like a nonspecific indefinite NP, in that it prefers not to undergo object shift’.

To sum up, contrastive stress on the object overrides the actvn effect, both in Icelandic and in Swedish and Danish. But it does not override it completely, at least not in Icelandic. The findings for Icelandic show us that there is in fact no constraint on all contrasted objects in Scandinavian languages to appear in the in situ position; in Icelandic pronominal objects with contrast seem to shift as readily as those with no contrast.

3.2 Contrast on another element

Contrast influences object placement in yet another way. A non-contrasted object that would otherwise be strongly dispreferred in the shifted position due to the actvn effect may shift quite felicitously, when there is contrast on another element in the clause, for instance the verb, the sentence adverbial or the subject. This is a phenomenon that has been observed for Icelandic by Diesing & Jelinek (Reference Diesing and Jelinek1993:24; see also Diesing Reference Diesing1997; Thráinsson Reference Thráinsson, Baltin and Collins2001:190, Reference Thráinsson2007:32) and for Swedish and Danish by Andréasson (Reference Andréasson2009, Reference Andréasson, Butt and Holloway King2010). The example in (13) is from Diesing (Reference Diesing1997:412).

1. (13)

   

In a context with no contrast, the actvn effect would lead to the indefinite NP object bækur ‘books’ in (13) appearing in the in situ position. With contrast on the the verb les ‘read’ this changes, and the object is allowed in the shifted position.

Let us now turn to Swedish and Danish. Recall from the above that the actvn effect leads to a Swedish pronominal object with a non-factive matrix verb, for instance tro ‘think/believe’, appearing in the in situ position. In the Swedish Parole corpus 90% of the object pronouns in sentences with tro are in situ (Andréasson Reference Andréasson, Butt and Holloway King2010); there are only eleven examples with a shifted object pronoun.Footnote ¹² In six of these sentences there is a clear contrast on something other than the object referent. In the Danish corpus KorpusDk a similar pattern emerges.Footnote ¹³ Several of the shifted det in examples with tro involve contrast on element other than the object referent.

In the Swedish example (14) from the Parole corpus, the verb is contrasted, just as in the Icelandic example in (13) above.

1. (14)

   

In (14) the antecedent of det ‘it’ is hon är en mördare ‘she is a murderer’. The speaker does not negate her being a murderer, quite the opposite, since the following sentence implies that s/he actually fears there might be some truth in this. Instead the negated domain is the attitude of tro (att hon är en mördare ‘thinking (that she is a murderer)’). This attitude is negated, and contrasted with the attitude of fearing the same thing, which is expressed explicitly in the following sentence, Jag fruktar det ‘I fear it’.

Similar examples can be found in Danish. The example in (15) is from a novel.Footnote ¹⁴

1. (15)

   

In the context of example (15) the question if Ann thinks that Peto will be asking her father for her hand is uttered. In Ann's answer the verb tro is contrasted with ved in the following clause. The author has even marked the prosodic prominence that falls on the verb tro with italics.

There are examples both in the Swedish and in the Danish corpus where elements other than the verb are contrasted. In (16) the contrast is on the subject andre ‘others’.

1. (16)

   

The context of (16) is a discussion on how children, during the formal operational stage of the development of their cognitive abilities, become able to relate diseases to external causes. In the example a young child is expressing this in words. The antecedent of det in (16) is the proposition de får kræft på grund af luftforurening eller kemikalier ‘they get cancer from air-pollution or chemicals’ in the immediately preceding sentence. In the sentence with the shifted pronominal object the subject andre ‘others’ contrasts with the subject of the preceding sentence, nogle folk ‘some people’.

As we have seen, contrastive stress on elements other than the object also seems to override the actvn effect in all the languages discussed here, but not in the same way as when the object is contrasted. In Icelandic all non-contrasted objects actually seem to avoid appearing in the in situ position when there is contrast on another element. In Swedish and Danish on the other hand not all objects seem willing to shift, even in these contexts it is only personal pronouns that shift. All other objects must still be in situ. This indicates that there must be some other factor that is involved in mainland Scandinavian object shift than accessibility and contrast. We will return to this in Section 4.6.

To sum up, the impact of contrast is neither that all contrasted objects appear in situ nor that all non-contrasted objects shift. One way to account for the effect of contrast on another element could be to say that the objects escape from a position where they otherwise may be misinterpreted as carrying contrastive focus, along the lines of Holmberg's (Reference Holmberg1986) original proposal for pronominal object shift. However, this does not explain why not all non-contrasted objects can shift to avoid being misinterpreted as contrasted. The fact that some contrasted objects are allowed to shift in Icelandic further complicates the escape analysis. In the following section I will suggest an analysis where the actvn effect, the impact of contrast and also to some extent information structure interact.

4.1 The not too hierarchical model

I adopt a simple model for describing Scandinavian clause structure, the not too hierarchical model, where the phrase structure in Swedish, Danish and Icelandic is flat in the area between a finite element and an optional VP, see example (17) and (18), and Figure 2. The ordering of the constituents in this flat area is not fixed, but falls out from a ranking of violable OT constraints.Footnote ¹⁵

Figure 2.

C-structures with VP and without any VP.

1. (17)

   

2. (18)

   

The claim in Andréasson (Reference Andréasson2007a, Reference Andréasson, Butt and Holloway Kingb) is that there is only evidence for one functional projection in Swedish, an FP.Footnote ¹⁶ The Fʹ hosts the finite verb in main clauses and the complementiser in subordinate clauses (Engdahl, Andréasson & Börjars Reference Engdahl, Andréasson and Börjars2004). The area including and following the finite verb (the Fʹ domain) is flat, and the word order in this local domain is determined by structural, semantic and pragmatic factors. In main clauses with a finite main verb, there is no independent evidence that there is a VP in Swedish (see Dalrymple Reference Dalrymple2001:52 on Icelandic). In clauses where the main verb is non-finite, however, constituent tests show that there is evidence for a VP, at least in Swedish (Andréasson Reference Andréasson2007a).

4.2 Constraints on accessibility

The actvn values will be expressed here as a family of so-called alignment constraints, where each constraint expresses a tendency for elements with a certain actvn value to appear as far to the left as possible in the Fʹ domain.Footnote ¹⁷ More generally an alignment constraint requires a certain element to be aligned in relation to the edge of a given domain, for example the daughter nodes of the same mother node (see e.g. the discussion on the notion of domains in Sells Reference Sells2001b). Here the domain that the alignment constraints relate to is Fʹ, as shown in Figure 2 above.

One example of this type of constraint is head-l, see (19) below, which requires a head to appear at the left edge of a domain. In the Fʹ domain of a declarative clause, the head is the finite verb.

1. (19) head-l

   Align L (head, local domain, edge), i.e. align the head with the left edge of the local domain.

head-l is assumed to be ranked higher than all the constraints discussed in this paper, reflecting the fact that verb second in Scandinavian declarative clauses outranks any pragmatic constraint on word order in this area of the clause.

The definition of the alignment constraints on different actvn values follows the model of actvn₀-l in (20).

1. (20) actvn₀-l

   Align L (actvn, local domain, edge), i.e. align elements with the actvn value 0 with the left edge of the local domain.

The internal ranking between the different actvn values given in (21) demonstrates the tendency for elements that are contextually given to appear earlier in a sentence, and for elements that are new to appear later.

1. (21) actvn₀-l » actvn₁-l » actvn₂-l » actvn₃-l

These actvn constraints compete with other pragmatic constraints on word order in the Fʹ domain. First we will consider the faithfulness constraint in (22). This constraint rewards modifiers preceding the domain they modify, icon-mod (from Andréasson Reference Andréasson2007a); icon stands for ‘iconic’ and mod stands for ‘modification’. The relative order between a modifier and the modified domain should reflect their semantic relationship; one element that modifies another should precede and c-command the modified element. In the examples in this paper, an iconic word order should reflect the fact that a sentence negation or a sentence adverbial modifies the clause.

1. (22) icon-mod

   Modifiers precede and c-command the domain modified.

In an Fʹ domain with a sentence negation, this constraint is maximally satisfied only if all other elements in the sentence follow the negation. However this is not grammatical in Scandinavian declaratives, see (23). This is illustrated by the head-l constraint, from (19) above, outranking icon-mod, see OT Tableau 1.

1. (23)

   

Tableau 1.

head-l vs. icon-mod.

In Tableau 1, the word order inte hörde (example (23b)) satisfies the constraint icon-mod in that the verb follows the modifier. Nevertheless, the ranking of head-l higher than icon-mod demonstrates that the winner is the word order where the finite verb is at the left end of the Fʹ domain. To simplify, I will in the following leave out the constraint head-l in all tableaux where the verb is not part of the competition, and only include candidates that fulfill this constraint.

We will now turn to constraints on object placement. As we will see in the following, it is the ranking of icon-mod in relation to the relevant actvn constraints that accounts for the facts about object shift in neutral contexts in Swedish, Danish and Icelandic. Example (24) and Tableau 2 illustrate that the shifted word order is preferred in Danish, when the object is a pronoun with the actvn value 0. This holds also for standard Swedish, see example (1) above, and for Icelandic, see example (8) above.Footnote ¹⁸

1. (24)

   

In Tableau 2icon-mod is violated twice for the shifted word order, since two elements, kender ‘know’ and hende ‘her’ precede the modifier ikke ‘not’. The in situ word order only violates this constraint once. Nevertheless, the referent of the pronoun hende is fully activated and since the constraint actvn₀-l outranks icon-mod, the violations of icon-mod do not affect the outcome of the competition. In the shifted word order only one element precedes the pronoun hende in the domain and this violates actvn₀-l once. The in situ word order gets two stars since both the verb and the negation precede the object pronoun. The shifted word order has the least violations of the highest ranked constraint, and it is the winner, marked by the pointing hand.

Tableau 2.

actvn₀-l vs. icon-mod in Danish (Swedish and Icelandic).

In (25) the antecedent is the entire proposition in the question and the pronoun det, has the actvn value 1.Footnote ¹⁹

1. (25)

   

In Tableau 3 it is the ranking of icon-mod higher than actvn₁-l (and all lower actvn constraints) that makes the in situ word order the winner.

Tableau 3.

icon-mod vs. actvn₁-L in Swedish (and Danish).

In Icelandic, icon-mod is ranked significantly lower than in Swedish/Danish. As we have seen, in Icelandic not only pronouns but also definite full NPs shift, see (26) (from (7b) above)). This is illustrated by the ranking of all the constraints actvn₀-l » actvn₁-l » actvn₂-l higher than icon-mod.

1. (26)

   

Also in this case icon-mod is violated twice by the shifted word order. Nevertheless, the ranking of actvn₂-l higher than icon-mod makes the shifted word order win the competition.

When the Icelandic NP is indefinite as in (27) (from (6a)), the in situ word order is the best in a neutral context.

1. (27)

   

In Tableau 5 it is icon-mod that gets to pick the winner, since the active constraint on givennness, actvn₃-l, is ranked lower than icon-mod to account for the fact that the in situ word order is preferred for indefinite NP objects.

Tableau 4.

actvn₂-l vs. icon-mod in Icelandic.

Tableau 5.

actvn₃-l vs. icon-mod in Icelandic.

4.3 Constraints on information status

As we saw in example (7) above, definite NPs may appear preceding or following a negation in Icelandic. Here, the ranking of icon-mod in relation to the actvn constraints does not give us the winner. However, the distribution is not random. Thráinsson (Reference Thráinsson2007:76) notes that the information status of the objects may play a role here (see also Broekhuis Reference Broekhuis2000 for a similar account, and Diesing Reference Diesing1997 for a different opinion).Footnote ²⁰

In this paper, I make a distinction between accessibility, see above, and the informative intentions a writer has with a sentence. In written text, where prosody is not present, word order is the main tool for making these intentions clear to a reader, a syntactic information packaging (see Vallduví & Engdahl Reference Vallduví and Engdahl1996). Information that a writer or speaker is primarily intending to add to what is already under discussion is packaged as the information status rheme. The parts of the statement that connect the rheme to what is under discussion is packaged as the information status ground, see (28) (Andréasson Reference Andréasson2007a).Footnote ²¹ The general constraint on information status is resp-i, see (29) (Andréasson Reference Andréasson2007a).

1. (28)

   1. a. rheme: The information in a statement that is intended to increase the listener's knowledge.

   2. b. ground: Constituents that relate the rheme to questions the speaker assumes are under discussion

2. (29) resp-i

   The linear order of constituents respect the information principle: i.e. ground < rheme.

resp-i is a faithfulness constraint rewarding an ordering of constituents in a clause based on their information status. This constraint is related to the typological tendency for ground material to appear earlier in a clause than rhematic elements, whenever structural constraints allow this.

In example (30) below (from Thráinsson Reference Thráinsson2007:76), the book title Stríð og frið, a definite description with actvn value 2, has not been mentioned in the context. Instead it is Jón's unknown regular activities during his vacation that are under discussion, and the rheme of the sentence is his reading War and Peace. Hence the object is part of the rheme, as shown by the attribute–value matrix in (31) which gives the LFG i(nformation)-structure that constitutes part of the input for the OT competition (see Andréasson Reference Andréasson, Butt and Holloway King2007b).Footnote ²²

1. (30)

   

2. (31)

   

In this context the habit expressed by the adverb alltaf ‘always’ is implied in the question. In the shifted word order, two elements representing the rheme, les ‘reads’ and Stríð og frið, precede this ground element, and the constraint resp-i is violated twice. The in situ word order only gets one star, since there is only one rhematic element that precedes alltaf. The ranking of resp-i higher than actvn₂-l illustrates that the in situ word order is the winner, when this particular information structure is part of the input.

In example (32) below (from Thráinsson Reference Thráinsson2007:76), on the other hand, the book Stríð og frið is mentioned in the context. The question whether John knows this book is under discussion, i.e. a possible but yet unknown relation between Jón and War and Peace constitutes the ground part of the sentence. The answer is affirmative: Já ‘yes’, Jón knows this book. And then the speaker continues on the same subject and adds that Jón always reads this book during his vacation, i.e. the rheme of this sentence fills in the blank; the evidence for his familiarity with War and Peace is his habitual reading of the book during his vacation.

1. (32)

   

2. (33)

   

In this context the in situ word order is marked.Footnote ²³ This is modelled by the constraint on information status, resp-i being ranked higher than icon-mod. Instead, the shifted word order as in (32a) is preferred, where the ground element Stríð og frið precedes the rhematic elements alltaf ‘always’ and í fríinu sínu ‘during his vacation’. The rhematic verb, as usual, is constrained by the highly ranked head-l to appear at the left edge of the Fʹ domain, and this violates resp-i once. In the in situ candidate, resp-i is violated twice, since two rhematic elements precede Stríð og frið.

4.4 Constraints on contrastive focus

Let us now turn to the effects of contrastive focus. Both Andréasson (Reference Andréasson2007a, for Swedish) and Engels (Reference Engels2012, for German, English and French) discuss how sentence adverbials ‘multitask’ in sentences with a contrastive focus (Engels’ focus).Footnote ²⁴ On the one hand they retain their semantic function as proposition modifiers, on the other hand the presence of a contrastive focus triggers a sentence adverbial to play the part of a focus operator.

As we saw in Section 3 above, contrastive focus affects object placement in two ways. It makes contrasted objects remain in situ, and it makes non-contrasted objects shift when there is contrast on another element in the clause. Engels (Reference Engels2012:112) proposes two separate constraints to deal with these effects of contrast on object positions in German, see (34).

1. (34)

   

Engels’ (Reference Engels2012) adverb < +focus covers the tendency for contrastively focused (in the following +foc or focus domain) objects to appear to the right of a focus operator (in the following f-op), here a sentence adverbial, in examples like (10) and (11) above. Engels’ –focus < adverb on the other hand covers the tendency for non-focused elements (in the following –foc) to appear preceding a sentence adverbial in sentences like (13)–(16) above. Examples (35) and (36) below are Engels’ (Reference Engels2012) (3.17a) and (3.18a).

1. (35)

   

2. (36)

   

In both these examples it is the presence of a +foc element that triggers the word order where the complements precede the sentence adverbial. In (35) it is the verb that is +foc and in (36) it is the subject. In both examples the sentence adverbial acts as the f-op.

In this paper I will make use of Engels’ (Reference Engels2012) constraints, with some adjustments, see (37) below. The original adverb < +focus constraint makes the right predictions in its present form; the order sentence adverbial preceding +foc is preferred. The constraint –focus < adverb, however, does not explicitly mention an input including a +foc element. Hence, it would constrain all non-focused elements to appear preceding a sentence adverbial, whether there is a +foc element in the sentence or not. By using a notion focus operator (f-op) instead of adverb, I ensure that both constraints are vacuously fulfilled in sentences with no elements with +foc. Referring to an f-op in the constraint also opens up for using the same or a similar constraint for other languages where elements other than sentence adverbials act as an f-op.

In (37), I also adjust Engels’ (Reference Engels2012) constraints to the theoretical assumptions in my analysis. The non-hierarchical model has a flat Fʹ domain; hence the notion of c-command will not suffice to represent the linearisation of the elements in the constraint. Therefore this constraint will also refer to precedence.

1. (37)

   

Example (38) below (repeated from (10)) shows that a pronominal object with contrast interpretation is required to appear in the in situ position in Danish. The same holds for Swedish. Since there is no context provided, the i-structure is underspecified for information status. Consequently the i-structure input in (39) just specifies the focus domain with the sentence adverbial acting as an f-op.

1. (38)

   

2. (39)

   

The constraint f-op < +foc is ranked higher than all actvn constraints in Swedish and Danish, and therefore this constraint settles the competition in Tableau 8. The shifted word order is the best candidate when it comes to actvn, but the in situ word order is the best candidate when it comes to contrast on the object. The constraint f-op < +foc is violated once in the shifted word order, and the winner is the in situ word order. In a sentence with no +foc this constraint would be vacuously fulfilled and icon-mod would pick the winner.

Let us now turn to contrast in Icelandic. As mentioned above, Collins & Thráinsson (Reference Collins and Thráinsson1996:406) state that contrasted definite NP objects in Icelandic must be in the in situ position, see (40). The ranking for Icelandic in Tableau 9 below accounts for this.

1. (40)

   

2. (41)

   

In Tableau 9 the actvn constraint actvn₂-l is violated twice by the in situ word order ekki bækurnar ‘not the books’, with contrast on bækurnar. However the ranking f-op < +foc » actvn₂-l makes the in situ word order the winner, since this word order does not violate f-op < +foc by placing a contrasted element in position other than directly to the right of the focus operator, the negation. Also actvn₃-l is ranked lower than f-op < +foc, which matches the fact that indefinite objects remain in situ when contrasted, just as they do in a neutral context with no contrast.

Let us turn to the facts about contrasted pronominal objects to see where f-op < +foc is to be ranked in relation to actvn₀-l in Icelandic. Example (42) from Thráinsson (Reference Thráinsson2007:32, 67) (repeated from (11) and (12) above) shows that pronouns with the actvn value 0 may appear shifted or in situ in Icelandic.

1. (42)

   

2. (43)

   

In Tableau 10 the constraint actvn₀-l is violated twice for the in situ word order while f-op < +foc is violated once for the shifted word order. The dashed lines between the constraints indicate that the competition between the two candidates is not fixed. The ranking may be actvn₀-l » f-op > +foc with the shifted word order as the winner, as illustrated in the tableau, or f-op < +foc » actvn₀-l with the in situ word order as the winner. It is unclear which factor decides when a contrasted object pronoun is in situ and when it is preferred to be shifted, but the most intriguing of these positions is without doubt the shifted one, since it makes it clear that not all contrasted pronominal objects have to follow a sentence adverbial (if present) in Scandinavian languages. Furthermore, I have at this point no clear data for elements with actvn value 1 in Icelandic, therefore the ranking for f-op in relation to actvn₁-l cannot be settled here either.Footnote ²⁵ I will have to leave these two questions for future research, and this is marked with dashed lines in the tableau.

As we saw in the previous section, there is also a tendency for non-contrasted elements to be dispreferred in the in situ position, in sentences where there is contrast on another element. This tendency is captured by the constraint –foc < f-op, see (37b) above.

The constraint –foc < f-op penalises non-contrasted elements appearing in the in situ position, but only when the i-structure input contains a focus domain and an f-op. When there is no contrastive focus, and hence no f-op, in the input, this constraint is vacuously fulfilled and the ranking of other constraints decide the competition.

In the Icelandic sentence from Diesing (Reference Diesing1997:412) in (44) (repeated from (13)) it is the verb that is contrasted.

1. (44)

   

2. (45)

   

In Icelandic the constraint –foc < f-op is ranked higher than both icon-mod and actvn₃-l, and it is violated by an in situ word order, where a non-contrasted element follows the f-op. Instead, the shifted word order is the winner. For Icelandic, we know that –foc < f-op must be ranked higher than icon-mod to allow shifted indefinite NPs in these contexts. It is however not settled where this constraint is ranked in relation to the constraints higher in the hierarchy and notably in relation to resp-i.

In Swedish and Danish it is objects with the actvn value 1 that unexpectedly appear in the shifted position when there is contrastive focus on another element. The constraint –foc < f-op must also in these languages be ranked higher than icon-mod, which here means that it outranks actvn₁-l and all other lower ranked actvn constraints. I repeat the Swedish example (14) here as (46).

1. (46)

   

2. (47)

   

In Tableau 12, –foc < f-op is violated once by the in situ word order TROR inte det ‘not it’. The shifted word order on the other hand does not violate this constraint and TROR det inte is the winner.Footnote ²⁶ The competition between –foc < f-op and actvn₀-l on the other hand cannot be settled at this time, since both reward a shifted word order.

4.5 Comparing Swedish/Danish and Icelandic

In Sections 4.2–4.4 I have presented constraints on linguistic factors that affect object positions in the Fʹ domain in Swedish/Danish on the one hand and Icelandic on the other. In this section I recapitulate the similarities and differences between the rankings.

1. (48)

   

Looking at the partial ranking in (48) we find that there are both similarities and differences between Swedish/Danish and Icelandic. The highest ranked constraint for both varieties is head-l, which ensures that the head of the Fʹ domain, i.e. the finite verb, precede all other elements in this domain (Tableau 1 above). In Icelandic it has not been settled where f-op < +foc is ranked in relation to the two highest actvn constraints (Tableau 10 above), whereas it has been shown that this constraint is ranked higher than all actvn constraints in Swedish/Danish (Tableau 8). The vertical bars in the ranking for Icelandic represent the competition not being settled.

The icon-mod constraint is also ranked differently for the two varieties. For Swedish/Danish this constraint is ranked high, to account for the fact that only objects with the highest actvn value shift in neutral contexts. In Icelandic, this constraint is ranked lower, between actvn₂-l and actvn₃-l to account for elements with actvn_0–2 shifting in neutral contexts (Tableaux 4 and 5). The constraint on information structure, resp-i, is ranked higher than actvn₂-l to account for rhematic definite objects appearing in the in situ position in Icelandic (Tableaux 6 and 7). The role of resp-i in relation to object positions in Swedish and Danish has not been discussed here, but see Ørsnes (Reference Ørsnes2013) for an account for Danish.

Tableau 6.

resp-i vs. actvn₂-l, Icelandic I.

Tableau 7.

resp-i vs. actvn₂-l, Icelandic II.

Tableau 8.

Contrasted pronominal object, Danish (and Swedish).

Tableau 9.

Contrasted definite NP object, Icelandic.

Tableau 10.

Contrasted pronominal object, Icelandic.

The constraint –foc < f-op is not included in (48). The ranking of this constraint is not completely set, but (49) shows its ranking relative to icon-mod.

1. (49) Swedish/Danish and Icelandic

   –foc < f-op » . . . » icon-mod

Tableaux 11 and 12 show that –foc < f-op must be ranked higher than icon-mod both in Swedish/Danish and in Icelandic to account for the fact that objects that would otherwise appear in situ do shift when another element in the sentence is +foc. The dots in (49) represent any other constraint that may intervene between –foc < f-op and icon-mod.

Tableau 11.

Contrast on another element, Icelandic.

Tableau 12.

Contrast on another element, Swedish (and Danish).

This paper mainly deals with object positions relative to sentence adverbials in the Fʹ domain. To develop a more general OT analysis of word order in this domain, we need to include also subjects and grammatical relations. Furthermore, a complete account of object placement needs to take the initial position into account but this is beyond the scope of this article, see Engdahl et al. (Reference Engdahl, Andréasson and Börjars2004) and Andréasson (Reference Andréasson2007a) on the word order in the Fʹdomain, and Ørsnes (2013) for an account including the initial position. There is, however, a problem with the suggested ranking of –foc < f-op in Swedish/Danish which I will discuss in the next section.

4.6 The need for a constraint on prosody

Tableau 12 above, shows that the constraint –foc < f-op must be ranked higher than the actvn constraints with values 1 and higher in Swedish and Danish. Unfortunately, such a ranking would also allow all nominal objects to shift in a sentence with contrastive focus on some element other than the object. This is not grammatical in Swedish and Danish. In the following, I will draft a possible solution to this.

Josefsson (Reference Josefsson2010) presents an analysis which takes into account syntactic, information structure and prosodic aspects of pronominal object shift in Swedish. Josefsson focusses on the prosodic features and her main purpose is to investigate to what extent it is optional for unstressed pronominal objects to appear in situ, in syntactic environments where object shift is possible in Swedish. Josefsson suggests that shifted objects form a prosodic word with the element to their left, whether it is the finite verb, a subject NP or an adverb. This word unit has the Swedish word accent 1, where there is stress on the first syllable and any following syllables are unstressed. In Josefsson's investigation the informants were asked to read out test sentences, with no stress on the object pronoun, and judge their acceptability. The informants seemed to accept the two object pronouns, henne ‘her’ and honom ‘him’ in situ more readily than the monosyllabic pronouns. Josefsson proposes that this may be due to henne and honom being disyllabic, and hence prosodically ‘heavier’ than the monosyllabic pronouns.

Andréasson (Reference Andréasson, Butt and Holloway King2008, Reference Andréasson, Butt and Holloway King2010) shows that non-contrasted henne and honom appear in situ only to a very limited extent in writing. Only about 5% of the henne/honom from written sources appear in situ without a clear contrast in the context. Bentzen et al. (Reference Bentzen, Anderssen and Waldmann2013) on the other hand have found that as much as 36% of all Swedish non-contrasted object pronouns with NP antecedents appear in situ in written sources, and that henne/honom ‘her/him’ only represent three out of the total of 17 instances found, while 14 instances are monosyllabic. This discrepancy between the results of Andréasson, Josefsson and Bentzen et al. is interesting. Bentzen et al. have worked with authentic spoken material and it seems that pronominal objects remain in situ to a greater extent in speech than in writing. However, Bentzen et al. have not had access to the sound files for the Swedish material, and to confirm their results, a more thorough prosodic analysis would have to be performed.

Nevertheless, we see a possible difference between spoken and written Swedish. Recall example (4) above, where the difference in pronunciation of the pronoun det was the only thing that indicated that the antecedent in (4a) was the computer game and in (4b) the act of Agnes buying a computer game. In writing, the answer would have been ambiguous.

In spoken language there is a possibility for prosodic marking of contrast and, as we have seen here, of accessibility. This prosodic marking is not available in writing. In written text we must instead use other means to avoid ambiguity. In example (4), a writer may have considered using an expression with more descriptive content than a pronoun to resolve the ambiguity. And – as we have seen – in a sentence where object shift is an option, the shifted word order may be used to syntactically mark a high actvn value as well as contrast on another element.

We know that only unstressed object pronouns are allowed in the shifted position in Swedish and Danish. Let us return in example (50) to the sentence in (46) above, now marked with prosodic information.

1. (50)

   

In Section 2, I showed that the difference between elements with the actvn value 0 and 1 in the examples discussed in this paper is prosodic and not lexical. actvn value 0 is unstressed, and value 1 is slightly stressed. In this example, the object pronoun det ‘it’ is embedded under the non-factive matrix verb tro ‘think’, which I take to mean that its actvn value is 1. In a sentence where det is in situ (see example (2) above), or for example in a sentence with no sentence adverbial, the pronoun would be slightly stressed. In (50) it is unstressed. It is not likely that contrast on another element has suddenly promoted the referent of this pronoun to actvn value 0, i.e. made the referent more accessible. Rather it seems that the pronoun has been de-stressed to be allowed in this position. Even if this needs to be investigated further on naturally occurring data, it does suggest that some objects actually get de-stressed regardless of their actvn value.

A highly ranked constraint on the de-stressing of objects that appear in the shifted position is one way to prevent elements other than pronouns from shifting in mainland Scandinavian. All nominal elements except pronouns carry some kind of word stress.Footnote ²⁷ This constraint must be ranked higher than –foc < f-op in mainland Scandinavian. In Icelandic such a prosodic constraint should be ranked low, at least lower than –foc < f-op and the other constraints discussed in this article. I will not attempt to include such a constraint in this analysis; instead I will leave this for future research.