1 Introduction

It is a well-known property of language that the surface structure does not always match the interpretation one-to-one. Moved elements, for instance, can in some cases be interpreted either in their derived position or in their base position (Postal 1974). Quantified subjects are a case in point, as they can often get both wide and narrow scope readings. In the raising context in (1), for instance, the existential quantifier can take scope either above or below the raising predicate.

  1. (1)
    figure a

On the wide scope reading of (1), there is a woman such that she is likely to buy some particular dress in question, while on the narrow scope reading, it is likely that some woman (or other) will buy that particular dress.

It is generally assumed that reconstruction of a phrase is only possible into positions that the phrase has occupied at some stage of the derivation (see e.g. Romero 1998; Fox 1999; Sportiche 2006a). It follows, then, that a subject that cannot be interpreted scopally in the position where it appears to be assigned its θ-role might never actually have appeared in that position (Chomsky 1993; see also Sportiche 2006b). Precisely this issue comes up in analyses of tough-constructions (TCs) (for discussion, see e.g. Fleisher 2013).

In English, TC subjects can only get a wide scope reading. Thus, despite the fact that the subject in (2) is thematically interpreted in the embedded object position, it does not reconstruct into this position for scope (see e.g. Postal 1974; Epstein 1989).

  1. (2)
    figure b

As indicated, the sentence in (2) has the reading that few girls x are such that it would be difficult for John to talk to x, but lacks the meaning on which it is difficult for him to talk to a small group of girls. The reconstruction data in (2) seems straightforward enough: the absence of a narrow scope reading of the subject can be taken as support for analyses that take the matrix subject to be base-generated in the surface position (see among others Chomsky 1977; Browning 1989; Rezac 2006; Fleisher 2013, 2015; Keine and Poole 2016), and, conversely, to be problematic for analyses that assume movement into the subject position (see among others Rosenbaum 1967; Postal 1971; Brody 1993; Hornstein 2001; Hicks 2009; Hartman 2011a,b).

Swedish is particularly interesting in this context because different types of TC behave in different ways with regard to scope reconstruction. A preliminary division can be made between adjectival TCs, (3a), and verbal (and nominal) TCs, (3b). The former are like their English counterparts in only allowing for a surface scope reading, while the latter allow for both readings:Footnote 1

  1. (3)
    figure c

The sentence in (3b) can mean either that few people are such that each one of them is easy for Johan to talk to (wide scope reading), or that it is easy for Johan to talk to a small group of people (narrow scope reading). The sentence in (3a), in contrast, can only mean that few people are such that each one of them is easy for Johan to talk to. To my knowledge, this difference in reconstruction ability of different types of TC in Swedish has not been discussed previously in the literature.Footnote 2

In fact, while the adjectival type of TC in (3a) has been examined to some extent before in Swedish, other types of TC (of which (3b) is just one example) have not received much attention. For instance, although Malmgren (1984) gives examples of some verbal TCs, his discussion is focused on adjectival ones. Similarly, Engdahl (2012) is concerned with properties of the non-reconstructing, adjectival type and concludes that TCs in Swedish have base-generated subjects. The contrast in reconstruction properties of different types of TC, (3b)–(3a), has thus not been noticed. If the ability to reconstruct indeed correlates with movement, as argued in, among others, Romero (1998), Fox (1999) and Sportiche (2006b), then the data in (3) seems to suggest two different derivations of TCs in Swedish. More precisely, those TCs that allow for reconstruction of the subject, (3b), would be derived via movement, while those that do not reconstruct, (3a), would involve a base-generated subject.

In this paper I will be concerned with the derivation of TCs in Swedish, but I will also look at correlations between morphological agreement and lack of reconstruction in Swedish more generally. Regarding TCs, I will argue against a solution with two derivations for these constructions and present evidence that both adjectival and verbal types are derived via movement from the embedded clause. Firstly, TCs behave uniformly when it comes to reflexive binding, indicating that the subject has moved. Secondly, those TCs that fail to reconstruct do so for independent reasons, I will argue, and not because they do not involve movement of the subject. Specifically, I will show that reconstruction is unavailable in Swedish more generally in contexts with morphological agreement between the subject and the predicate. The pattern observed for TCs, (3), thus holds also for passives and raising structures. It will be argued that morphological agreement arises from ϕ-feature agreement between the subject and predicate in Swedish, and that cases without a morphological reflex as well as cases with a default morphological form do not involve ϕ-feature agreement. A subject that has agreed in ϕ-features with its predicate is frozen in place for scope reconstruction purposes.

The paper is structured as follows: In Sect. 2, I begin by laying out the properties of TCs in Swedish, looking at predicate types, binding properties and scope reconstruction properties. As will be shown, both binding and scope provide evidence for movement into the subject position in TCs in Swedish. For the ability to reconstruct, however, it matters whether the predicate agrees morphologically with the subject or not. Section 3 therefore looks at morphological agreement in Swedish, both straight-forward and unexpected agreement patterns. I argue that common gender and plural agreement marking arise as a result of ϕ-feature agreement but that ϕ-feature agreement is absent in other subject-predicate relations in Swedish. When ϕ-feature agreement has taken place, the subject is frozen in place and necessarily gets wide (non-reconstructed) scope. I show that the proposed analysis of ϕ-agreement and scope freezing carries over also to passives and raising constructions. In Sect. 4, I turn to the syntactic derivation of TCs. I argue that TCs are derived via a combination of A′- and A-movement of the underlying object into the matrix subject position.

2 Properties of TCs in Swedish

In this section, I look at the properties of TCs in Swedish in some detail. As TCs in Swedish differ in their ability to reconstruct depending on their type of predicate, I begin, in Sect. 2.1, with an overview of the different predicate types found. After that I turn to expletive-drop constructions in Sect. 2.2, as such constructions superficially look like TCs, but are distinct from these. Sections 2.3 and 2.4 discuss binding and scope properties of TCs.

2.1 Predicate types

Swedish has a number of TC types that differ in the form of their predicate: adjectival in (4a), nominal in (4b) and verbal in (4c)–(4h).Footnote 3Footnote 4Footnote 5

  1. (4)
    figure d

Of the different TC types in (4), the adjectival and the verbal ones with gå (‘go’), (4c)–(4d), are particularly productive and can appear with almost any kind of infinitival complement.Footnote 6 The other types are more restricted.

There are structural differences between the TC types. In verbal TCs, the matrix verb forms an essential part of the tough-predicate and cannot be separated from its complement when a wh-question is formed. In adjectival TCs, the verb and the adjectival complement can be separated, as the verb is a copula and is not part of the tough-predicate.Footnote 7

  1. (5)
    figure h

In addition to syntactic differences, there are also morphological differences between the predicates. While adjectival tough-predicates show morphological agreement with the subject, nominal and verbal ones do not. I will return to this property in Sect. 3.

TCs are different from other, superficially similar adjectival constructions in that they have expletive counterparts. In this respect, they are similar to raising constructions. The subject of the TC thus has the same thematic interpretation as the embedded object in the expletive counterpart.Footnote 8 All TC types exemplified above have expletive counterparts, (6a)–(6h) below. Indeed, Hicks (2009), among others, takes the ability to alternate with expletive constructions to be a defining property of TCs.

  1. (6)
    figure i

The sentences in (6) are clear examples of expletive constructions. In the next section, we will see that expletive constructions can look identical to TCs under certain conditions. Such constructions need to be distinguished from TCs as they have other properties.

2.2 TCs and expletive-drop constructions

In Swedish, expletive subjects can be left out under certain conditions (for details, see Falk 1993; Engdahl 2010, 2012; Platzack 2012b, 2013). The nominal and verbal TCs in (4b)–(4h), above, are a case in point: they are structurally ambiguous, which can be shown by adding tag questions to them. The subject in the tag is a pronoun corresponding in number and gender to the subject of the main clause. In one structural variant, the first constituent (den här artikeln) is the syntactic subject, (7a), while in the other variant, there is an expletive subject that is not pronounced in the main clause but is pronounced in the tag, (7b).Footnote 9 The sentences in (7a)–(7b) illustrate this for the verbal go-TC, but the pattern holds for all of the TC types in (4b)–(4h).

  1. (7)
    figure k

In (7b), the first constituent (den här artikeln) is not a subject but a fronted object and the sentence is therefore not a TC.

Notably, only sentences with a verbal, nominal or neuter-marked adjectival predicate are ambiguous as in (7a)–(7b). With an adjectival predicate marked for common gender or plural, there is only one possible structure, (8), because common gender and plural-marked predicates are incompatible with expletive subjects (whether pronounced or dropped), (9).

  1. (8)
    figure l
  1. (9)
    figure m

Since structures with a dropped expletive subject are not TCs, it is important to make sure the sentences discussed below do not have a dropped expletive. The ambiguity in sentences like (4b)–(4h) arises because subjects in Swedish can appear either before or after the verb. If an adverbial is placed in the clause-initial position, the ambiguity disappears. In that case, the overt DP following the verb must be the subject, leaving no room for a dropped expletive:

  1. (10)
    figure n

With a clause-initial adverbial, the sentence in (10) is a clear case of a TC. This is also the case if the sentence is turned into a polar question. As in (10), the overt DP following the verb is unambiguously the subject:

  1. (11)
    figure o

In the rest of the paper, I will include clause-initial adverbials or use the polar question form in the otherwise ambiguous cases to make sure that the sentences in question are TCs rather than expletive constructions.Footnote 10

In the next two subsections, I turn to binding and scope reconstruction properties of TCs in Swedish. Such properties can be indicative of whether the constructions involve movement of the subject or not.Footnote 11

2.3 Reflexive binding

As expected, reflexives in subject position are ill-formed in Swedish when the subject is non-derived.

  1. (12)
    figure r

In constructions with derived subjects, as in (13) below, reflexives in the subject are accepted by some speakers, although others find these also ill-formed.Footnote 12

  1. (13)
    figure s

The sentences in (12)–(13) illustrate that, for those speakers that accept reflexives inside the subject, the reflexive is well-formed only if the subject has been moved from a position lower than the binding element. To other speakers, reflexives in subjects are ill-formed also in such cases, as mentioned above. This is also the case with the TC data below; Some speakers find the sentences marked or ill-formed, while others find them fully acceptable. I use % to indicate this.Footnote 13 With this in mind, reflexives should be possible to use as a one-way diagnostic for movement in TCs. If a reflexive is well-formed inside the subject, the subject must have moved from a position lower than the antecedent, but if it is not well-formed, no clear conclusions can be drawn. In this section, I present reflexive data that indicate that the subject has indeed moved, and, moreover that it has moved all the way from inside the embedded clause. The data are of three types: picture-noun contexts, contexts where the reflexive is not embedded under another noun, and ditransitives (see discussion in Poole et al. 2016).

The first type of data I look at is picture-noun contexts. As illustrated in the examples in (14), TCs in Swedish allow for a reflexive inside the subject, bound by an argument lower down in the structure. The examples in (14) have a reflexive pronoun and the ones in (15) have a reflexive possessive.

  1. (14)
    figure t
  1. (15)
    figure u

For those speakers who accept sentences like (14)–(15), the sentences become ill-formed when the för-PP containing the antecedent is removed, even in the presence of a salient potential discourse antecedent:Footnote 14Footnote 15

  1. (16)
    figure w
  1. (17)
    figure x

While varje fotograf (‘every photographer’) co-varies with the reflexive in (14)–(15), the same is not the case for fotograferna (‘the photographers’) in (16)–(17). The difference in well-formedness between these sentences indicates that the reflexives need to be bound and that the Experiencer för-PP can act as a binder.

The sentences in (16)–(17) show that picture-noun reflexives in Swedish need to be syntactically bound (in episodic contexts, see fn. 14). In English, on the other hand, picture-noun contexts have been shown not to require syntactic binding and are consequently not considered to be a reliable diagnostic for syntactic binding (e.g. Reinhart and Reuland 1991; Pollard and Sag 1992; Reinhart and Reuland 1993; Heycock 1995; Bhatt 2002).Footnote 16Footnote 17 While it is quite clear, thus, that Swedish is different from English with regard to picture-nouns, I nevertheless give examples in (18) below where the reflexive is not embedded under such nouns. Notably, even to speakers who find (18) degraded, there is a clear contrast between these sentences and the corresponding examples without the binder in the för-phrase.Footnote 18

  1. (18)
    figure ac

The examples in (19) show that the sentences are not well-formed in the absence of a syntactic binder:

  1. (19)
    figure ad

The examples in (14)–(19) show that reflexives are well-formed (to some speakers) inside the subject of a TC when there is an antecedent that can bind it (the för-PP). As pointed out by a reviewer though, this data is compatible with movement of the subject from a position lower than the antecedent för-PP but higher than the embedded clause. The data thus does not unambiguously show that the subjects have moved from inside the embedded clause. Poole et al. (2016) discuss this issue for English, using the ditransitive paradigm in (20) (their ex. (5)). As shown in (20c), English TCs do not allow for a pronoun inside a subject to be bound by something inside the embedded clause:

  1. (20)
    figure ae

For Swedish, a different pattern emerges: variable binding into the embedded clause is possible. In (21)–(22), the direct object (sin (rättmätiga) lön ‘his/her (rightful) salary’) is bound by the indirect object (varje anställd ‘every employee’) both in the expletive sentences, (21a), (22a), and the TCs, (21c), (22c). (The sentences in (21b) and (22b) are control sentences with no binding, as in (20b) above.)

  1. (21)
    figure af
  1. (22)
    figure ag

In (21)–(22), the reflexive appears in the direct object which is structurally lower than the indirect object binding it. If the reflexive appears in the indirect object instead, the sentence is no longer well-formed. The sentences in (23) below are expletive constructions illustrating this: (23a) is well-formed since the reflexive (sin baby ‘her baby’) appears in the direct object below the antecedent (varje mamma ‘every mum’), while (23b) is ill-formed, the reflexive (sin mamma ‘his/her mum’) appearing above its intended antecedent (varje baby ‘every baby’).

  1. (23)
    figure ah

The TC corresponding to the expletive construction in (23a) is well-formed, (24a). Forming a TC corresponding to the ill-formed sentence in (23b) is not possible, (24b). More specifically, the subject (sin mamma ‘his/her mum’) is interpreted as the thing given (thus corresponding to the direct object before movement) rather than the recipient (indirect object), as can be seen in the translation. Given the meaning of räcka (‘hand’), this is a very implausible reading.

  1. (24)
    figure ai

The ditransitive data shows that the reflexive cannot be merged in a position above its intended binder, i.e. it cannot be interpreted logophorically. Furthermore, since the reflexives in the subjects in (21c), (22c) and (24a) (sin (rättmätiga) lön ‘his/her (rightful) salary’ and sin baby ‘her baby,’ respectively) are bound by the indirect objects (varje anställd ‘every employee’ and varje mamma ‘every mum,’ respectively) in the embedded clauses, I conclude that the subject has moved from inside the embedded clause in these cases. The data looked at in this section show that for those speakers who accept reflexives in the subject position in TCs, the reflexives must have an antecedent syntactically binding it at some stage of the derivation.

2.4 Scope properties

When it comes to scope reconstruction properties, TCs in Swedish behave differently depending on what type of predicate they have. TCs with verbal and nominal predicates allow for the subject to reconstruct below the predicate, (25a)–(25d), and so do adjectival TCs whose predicate has neuter agreement marking, (26a)–(26b). Adjectival TCs with common gender or plural marking on the predicate, in contrast, do not allow for the subject to reconstruct, (27a)–(27b).Footnote 19

  1. (25)
    figure aj
  1. (26)
    figure ak
  1. (27)
    figure al

Possible readings for a few of the examples above are as follows: In (25a), an answer to the question on its surface scope reading (i.e. ‘Were few people such that each one of them was easy for Johan to talk to?’) could be ‘Yes, Anna was easy for him to talk to, and Pete was easy for him to talk to, and Kate too, but everyone else was difficult for him to talk to.’ An answer to the narrow scope question (i.e. ‘Was it easy for Johan to talk to a small group of people?) could be ‘Yes, any group consisting of fewer than five people was easy for him to talk to, but any group larger than that was a problem.’ Similarly in (26a), a possible answer to the wide scope reading question (‘Was one single piece of clothing such that it was difficult for Johan to buy?’) is ‘Yes, there was a specific jumper that he needed but which was difficult for him to buy, but no other pieces of clothing were difficult for him to buy.’ On the narrow scope reading (‘Was it difficult for Johan to buy just one piece of clothing?’) the answer could instead be ‘Yes, it was very difficult for him to restrict himself to buying just one piece of clothing as he really likes clothes shopping.’ Unlike the sentences in (25) and (26), the sentences in (27) only have the surface scope reading.

Before discussing why reconstruction is blocked in the common gender and plural cases, I will add a brief note on non-agreeing cases. For adjectival TC to block the reconstructed readings, as in (27) above, it is crucial that the predicate is marked in accordance with the common gender or plural subject. If it does not agree morphologically with the subject but is instead marked as neuter, reconstruction is possible (see also Malmgren 1984:103–104). Such non-agreeing constructions, illustrated in (28)–(29) below, are examples of expletive-drop constructions, as discussed in Sect. 2.2 above. In these sentences, an expletive can be inserted between the matrix verb and the adjective and they are therefore not TCs. With neuter agreement, the narrow scope reading is preferred over the surface reading. Since the surface scope reading does not seem to be completely ruled out, however, I have included it too in the translation but marked it as in the last line. The examples in (29) are from Malmgren 1984:104–105.Footnote 20Footnote 21

  1. (28)
    figure ao
  1. (29)
    figure ap

Leaving expletive-drop constructions aside, we have a situation where adjectival TCs (excluding neuter marked ones) differ from non-adjectival ones (verbal and nominal TCs) in their scope reconstruction properties.Footnote 22 On the assumption that the possibility of a narrow scope reading is dependent on previous movement and that the absence of such a reading therefore suggests absence of movement, the reconstruction data seemingly point in two different directions for the derivation of TCs in Swedish: via base-generation of the subject for the adjectival group (excluding neuter marked ones) and via movement of the subject for the non-adjectival group (verbal and nominal TCs). In the next section, however, I propose that lack of reconstruction can be independently explained in Swedish and that the lack of reconstruction therefore does not indicate lack of movement.

3 Predicative agreement and reconstruction in Swedish

In the following, I propose an analysis that ties the lack of reconstruction in adjectival TCs to the particular subject–predicate agreement taking place in these contexts. I begin by giving a short overview of predicative agreement in Swedish, before I turn to the analysis of the different agreement contexts in Sect. 3.1, and discuss why some of these contexts lead to scope freezing in Sect. 3.2. After that, I show that the analysis also extends to passives and raising constructions, in Sects. 3.3–3.4.

The obvious difference between adjectival and non-adjectival TCs is that in adjectival TCs, the predicate agrees morphologically with the subject while it does not in non-adjectival ones. In Swedish, adjectives in predicative position are marked for common or neuter gender in the singular, and for plural. The sentences in (30) exemplify all three markings. The sentences in (31) show that verbal and nominal predicates have the same morphological form irrespective of the gender and number of the subject.

  1. (30)
    figure aq
  1. (31)
    figure ar

The morphological form of the predicative adjective reflects properties of the subject that it agrees with. In regular cases, a common gender subject in the singular (e.g. artikeln ‘the paper’ in (30a)) will result in a common gender marked adjective, a neuter gender singular subject (e.g. utkastet ‘the draft’ in (30b)) will result in a neuter marked adjective and a plural subject (e.g. artiklarna ‘the papers’ in (30c)) in a plural marked adjective.

In some particular cases, the agreement pattern is less straight-forward (see Teleman et al. 1999:278–279, 288–289). For instance, subjects that contain co-ordinated clauses, (32a), or eventive nominals, (32b), do not result in plural agreement, but in neuter marking. These cases contrast with conjoined DPs of other kinds, (32c), which behave like plural nouns in co-occurring with plural agreement marking on the adjective (the examples in (32a) and (32c) are from Josefsson 2006:1347):

  1. (32)
    figure as

Although the subjects in (32a)–(32b) refer to more than one event or state of affairs, the predicative adjectives are not marked for plural.

Another unexpected pattern is found with bare nominals, which typically refer to mass entities or substances, as discussed by Josefsson (2006). The nouns senap (‘mustard’), grädde (‘cream’) and mjölk (‘milk’) are common gender nouns, (33a). When appearing in bare noun contexts, however, they give rise to neuter marking on the predicative adjective, whether they appear single or conjoined. Sentences of this type are similar to sentences like (34), which are referred to as “Pancake sentences” by Josefsson (2010, 2014). These have bare plural subjects rather than bare singular ones. The examples in (33b)–(33c) are from Josefsson (2006:1347–1348), and (34) from Josefsson (2014:62).

  1. (33)
    figure at
  1. (34)
    figure au

Sentences like (32)–(34) thus all exhibit an apparent mismatch between the subject and predicative adjective in number or gender.

To sum up, common gender marking on the predicative adjective is restricted to contexts with non-bare common gender singular subjects and plural marking is restricted to contexts with subjects in the plural that are neither clauses nor eventive nominals. Neuter marking, on the other hand, is found in contexts with neuter gender singular subjects as well as in contexts with bare subjects of either gender, eventive nominals and clauses appearing on their own or conjoined.

3.1 Gender and number agreement

Descriptively, two sets of features are relevant for subject-predicate agreement in Swedish: gender and number. The subject-predicate agreement patterns observed in the section above can thus be represented by features in the following manner for regular nouns:

  1. (35)
    figure av

While the feature set-up in (35) captures the agreement patterns for the regular cases, the specification for neuter needs to be revised or complemented to account for the less straight-forward cases discussed above. As we saw, clauses and eventive nominals do not result in plural agreement when conjoined. Example (36) below shows that they also resist plural forms. More precisely, if they can be pluralized at all, they lose their event meaning in the plural. The nominal in (36a) can thus have either an ‘event’ reading or a ‘thing’ reading, but when pluralized, (36b), only the ‘thing’ reading survives (examples from Josefsson 2006:1350):

  1. (36)
    figure aw

Building on Grimshaw (1990), Josefsson argues that the inability of clauses and eventive nominals to pluralize is due to their lack of a number feature (2006:1349). When these appear as subjects, they result in neuter marking on the adjective. Lacking a number feature, the reduced feature set-up for these cases would thus be the following:

  1. (37)
    figure ax

On closer inspection, it is clear that the feature set-ups given in (35) and (37) include some redundancies. More precisely, the number specification for the non-plural cases, (35a)–(35b), is not necessary to distinguish between the forms. If number is not present on these, the remaining feature is Gender. In the set-ups given, Gender is either +common or +neuter. This specification can also be simplified so as to include only one of these. I am going to propose that only common gender is present as a feature in the feature grid. That means that neuter marking is the result when there is absence of agreement. The neuter form is a default form in the sense of Preminger (2009). That is, the predicate has features that do not get valued in an agree relation, and therefore get default values.Footnote 23 The revised set-ups are as follows:

  1. (38)
    figure az

The reduced grids allow us to unify all cases of neuter. This is a welcome result since neuter gender contexts behave uniformly and are indeed different from common gender and plural ones when it comes to scope reconstruction, as discussed in Sect. 2.4 above.

When simplifying the feature grids above, we removed the Number feature from the non-plural cases.Footnote 24 Although singular DPs thus lack a Num feature, they are of course able to appear with the articles en (‘a/one’ common gender) or ett (‘a/one’ neuter gender). On my analysis, en and ett are indefinite markers, spelling out Definiteness features, rather than Number features. Definiteness is also involved in the difference between (33a) and (33b)–(33c) above. (33b)–(33c) lack a Definiteness projection.

Unlike adjectival predicates, which have ϕ-features valued in the agreement process, verbal predicates lack such features altogether in Swedish. Irrespective of type of subject, these predicates therefore look the same (see e.g. (31b) above).Footnote 25

  1. (39)
    figure bb

Verbal predicates come with an EPP feature, ensuring that a DP moves into their Specifier position.

3.2 ϕ-feature agreement and scope freezing

Let’s now return to adjectival predicates. While common gender and plural marking on the adjective is the result of ϕ-feature checking, neuter marking arises as a default form in the absence of agreement. Recall from Sect. 2.4 that these adjectival contexts also differ in their reconstruction abilities. Reconstruction is blocked in adjectival TCs with common and plural agreement, (27), but it is possible in non-adjectival contexts, (25), as well as in adjectival contexts with neuter agreement, (26):

  1. (27)
    figure bc
  1. (25)
    figure bd
  1. (26)
    figure be

I propose that the crucial difference between the contexts with common gender or plural agreement and all other contexts is the ϕ-feature agreement. When the subject and predicate agree in ϕ-features, the subject is frozen in place for scope purposes. When there is no ϕ-feature agreement between the subject and the predicate, the subject can, in principle, be interpreted either in its derived, surface, position, or in its base position (or any of its intermediate positions). Notably, ϕ-feature agreement with its resulting morphological marking is the exceptional case in Swedish as it is restricted to a subset of the adjectival contexts and does not take place in verbal contexts. Since scope freezing co-occurs with ϕ-feature agreement, scope freezing is also the exceptional case.

There are cross-linguistic differences in what makes reconstruction (un)available in languages. For Swedish I propose that it is connected to ϕ-feature agreement, as we have seen. In English, the situation is different. Verbal predicates are morphologically marked for third person singular, indicating that there is ϕ-feature agreement between the subject and the verbal predicate. Unlike in Swedish, this ϕ-feature agreement does not correlate with scope freezing in general. However, Sauerland and Elbourne (2002) report that in British English, collective nouns can take either plural or singular verb agreement, (40a)–(40b), but that with plural agreement, the subject obligatorily takes wide scope, (40b) (examples from Sauerland and Elbourne 2002:288) (cf. the examples in (29) from Malmgren 1984:103–104):

  1. (40)
    figure bf

On the wide scope reading, available in both (40a) and (40b), there is a northern team, such that it is likely to be in the final, while on the narrow scope reading, which is only available in (40a), it is likely that some northern team (or other) will be in the final. Sauerland and Elbourne (2002) argue that there are different features present in the two sentences in (40). Collective nouns like team have both an ordinary Number feature, and a Mereology feature,Footnote 26 but the Mereology feature is only optionally present on T: it is absent in (40a) but present in (40b). Agreement with the Mereology feature leads to scope freezing, while agreement with the Number feature does not. In this way, Mereology agreement in British English has similar effects as ϕ-agreement in Swedish. The reason for the scope freezing effect in the British English cases is the following, on Sauerland and Elbourne’s analysis: the Mereology feature must be checked in the overt syntax unlike the Number feature, which can also be checked at LF. When features are checked in the syntax, syntactic movement follows, whereas when features are checked only at LF, movement takes place at PF. Crucially, it is the syntactic positions that determine scope relations. If only a surface scope reading is available, the particular features involved have been checked in the syntax. If both readings are possible, the relevant features have been checked either in the syntax or at LF. To Sauerland and Elbourne, thus, the scope interpretation follows straight-forwardly from the syntactic structure. On the other hand, they need to allow for certain features to have the dual possibility of being checked either syntactically or post-syntactically.

To account for the Swedish TC data, I propose an analysis along the lines in Sauerland and Elbourne (2002). Thus, in Swedish, ϕ-features are obligatorily checked in the syntax and the ensuing movement is syntactic. This is the structure interpreted at LF, accounting for the obligatory surface scope reading. When there is no ϕ-feature checking and movement is purely EPP-driven, movement takes place at PF. In this case, the subject can either be interpreted in its syntactic position, i.e. get narrow scope, or it can Quantifier Raise to a higher position at LF, and then get wide scope. Since the reconstructed reading is not achieved via lowering or some other undoing operation on this analysis, it is not necessary to stipulate that such an operation is available only when no ϕ-feature agreement takes place. Instead the scope readings uniformly reflect the syntactic structure, or are achieved via Quantifier Raising at LF if a second, higher scope reading is available.Footnote 27

To sum up, subject-predicate agreement that involves ϕ-features results in scope freezing in Swedish. With common gender and plural agreement on the predicate, the subject can only get a surface scope reading, but with neuter gender and with verbal or nominal predicates both readings are possible because these do not involve ϕ-feature agreement. This analysis makes an empirical prediction: also in non-TC contexts involving movement, reconstruction should be blocked when the predicate is marked for common gender or plural, but should not be blocked otherwise. Examples of such structures are passives and raising constructions.

3.3 Passives

In Swedish, there are two types of passive: morphological and periphrastic. Morphological passives are verbal. They are formed by means of a verbal morpheme that attaches outside any tense morphology. As with other verb forms in Swedish, there is no agreement marking.

  1. (41)
    figure bh

Periphrastic passives, in contrast, are formally adjectival, in the sense that they make use of an auxiliary (vara ‘be’ or bli ‘become’) and a past participle that agrees morphologically in gender and number with the subject.

  1. (42)
    figure bi

Given that periphrastic passives involve subject-predicate agreement, just like predicative adjectives, but morphological ones do not as they are verbal, the two types of passive provide a good testing ground for the correlation between agreement and lack of reconstruction that I have proposed above.

An examination of passives reveals the following general pattern: morphological passives allow for reconstructed readings, (43a)–(43c), while periphrastic ones do not if the participle is marked for common gender or plural, (44a) and (44c). With neuter gender agreement, (44b), the reconstructed reading is possible although the surface scope reading seems to be preferred.

  1. (43)
    figure bj
  1. (44)
    figure bk

The sentence in (43a) is compatible with a situation where there was one particular abbreviation (e.g. MP for ‘Member of Parliament’) that was accepted by the publisher (maybe because all the other ones were not generally used or transparent enough), as well as a situation where the publisher accepted at most one abbreviation, irrespective of what it was (maybe because they had a policy against using abbreviations). In (44a), in contrast, only the first reading is available. The same contrast can be seen between (43c) and (44c). For (43b) and (44b), both readings are possible (although the surface reading is preferred in (44b)).

As pointed out by a reviewer, the reconstruction pattern for passives is complicated by the fact that stress can play a role. Essentially, stress can have the effect of strongly favouring one reading over another. To some but not all speakers, focus can also bring out a reading that is not otherwise available. Under the assumption that Focus structure is operative at LF, its interaction with scope readings comes as no surprise as scope relations too are determined at LF. This interaction deserves a full investigation, but it falls somewhat outside the scope of the present paper. For some relevant data, see the Appendix.

As we have seen, the basic pattern that emerges for passives is similar to what can be observed in TCs. Morphological passives allow for both surface and narrow scope readings of their subjects, while periphrastic passives only allow for the surface scope reading when the participle is marked for common gender or plural. Periphrastic passives with neuter gender agreement are more flexible in often allowing for both readings. Before turning to the derivation of TCs in Swedish in Sect. 4, I will briefly look at raising structures, another context with movement into the subject position.

3.4 Raising

Subject-to-subject raising constructions are another instance where we could test the prediction that ϕ-feature agreement leads to scope freezing. However, while minimal pairs of adjectival and verbal passives can be formed relatively freely in Swedish, this is generally not possible in the case of raising structures, as there are only a few adjectival raising predicates, the majority instead being verbal. Bearing this in mind, raising structures still exhibit a similar reconstruction pattern as the one in TCs. That is, quantified subjects can reconstruct below the raising predicate if the predicate is verbal, (45a), but not if the predicate is adjectival, (46a).Footnote 28

  1. (45)
    figure bl
  1. (46)
    figure bm

The sentence in (45a) has two readings. It can either mean that there are two particular pupils, e.g. Mary and Tom, who seem to absent. It can also mean that from the total number of pupils, there seem to be two missing, irrespective of who they are. In (46a), in contrast, the only available reading is the one on which two particular pupils, e.g. Mary and Tom, are required to be present. The reading on which there have to be (any) two pupils present is thus unavailable. For the expletive counterparts, the sentence in (45b) allows for both readings, while the sentence in (46b) strongly favours the narrow scope reading.

Thus, although adjectival and verbal raising constructions do not form minimal pairs, the pattern observed for TCs largely carries over to raising constructions.

4 The derivation of TCs in Swedish

In the sections above, I presented evidence from reflexive binding and scope reconstruction indicating movement into the matrix subject position of TCs in Swedish. In the following sections, I propose that TCs in Swedish are derived via movement of the embedded object in two steps: an initial A′-movement step in the embedded clause, followed by A-movement into the matrix subject position. Section 4.1 lays out the evidence for the initial A′-step. In Sect. 4.2 I discuss problems encountered when applying previous A′-movement analyses of English TCs to Swedish ones. In 4.3 I instead suggest a long-distance A′–A movement analysis of TCs in Swedish, placing them in between the English and German types of TC.

4.1 A′-movement in the embedded clause

The fundamental question for TCs is how the subject is linked to the embedded object position where it gets its thematic interpretation. As argued for English (Chomsky 1977), I suggest that the subject of the TC in Swedish is linked to the object position via an A′-step in the embedded clause (see also Engdahl 2012). Evidence for this comes from data like (47), where arguments (Lisa, Johanna, Pelle) appear between the matrix subject and the object position.

  1. (47)
    figure bn

The well-formedness of sentences like (47) indicates that the subject and the object in TCs in Swedish are linked via an A′-chain rather than an A-chain, since A′-chains are not sensitive to intervening arguments (Chomsky 1977). In this regard, Swedish TCs are like their English counterparts, but different from German TCs, as discussed in Wurmbrand (2003).

  1. (48)
    figure bo

Wurmbrand (2003) argues that sentences like (48a) are ill-formed in German because German TCs involve A-movement all the way from the embedded clause. A-movement is possible since the embedded clause has a reduced clause structure and lacks both the subject position and the head associated with accusative case.Footnote 29 In this way, issues associated with early raising analyses of TCs (Rosenbaum 1967; Postal 1971) do not arise. As will be discussed in Sect. 4.3 below, Swedish TCs are also likely to have a reduced embedded clause structure, although, unlike in German, they involve A′-movement rather than A-movement.

Further support for an A′-movement step in the embedded clause comes from the observation that TCs in Swedish license parasitic gaps, (49a)–(49c). Crucially, parasitic gaps are not licensed in structures involving only A-relations, such as passives, (49d):

  1. (49)
    figure bp

Taken together, these observations suggest that TCs in Swedish involve A′-movement in the embedded clause. In the next section I look at two types of analysis that have been proposed for English TC, both involving A′-movement of an Operator in the embedded clause. Both types are problematic, however, and in Sect. 4.3 I therefore propose a different analysis for Swedish TCs.

4.2 The matrix subject: Base-generation or smuggling

There are different ways in which an A′-moved element could be linked to the matrix subject. Broadly speaking, A′-analyses of TCs either assume that this is achieved through further movement into the subject position (movement analyses) or via a base-generated subject that enters into an interpretive relation of some kind with the A′-moved element (an Operator) in the embedded clause (base-generation analyses).

The base-generation analyses have in common that the subject of the TC is merged in a non-theta position, matrix Spec,TP (e.g. Chomsky 1977; Browning 1989; Heycock 1994; Rezac 2006; Engdahl 2012; Fleisher 2015; Keine and Poole 2016). Via a relation (of some kind) between the subject and the moved element, a link between these two is formed and, as a result, the subject and the Operator share the same theta-role.

  1. (50)

    The booksare easy [OPPRO to read t]

Base-generation analyses predict that neither subjects with reflexives bound in the embedded clause, nor narrow scope readings of quantified subjects should be possible in TCs. As we have seen, however, neither of these predictions is borne out for Swedish. Analyses that assume A′-movement to be followed by another movement step into the matrix subject position fare better in this respect. Like base-generation analyses, movement analyses can also account for the fact that intervening arguments are allowed and parasitic gaps are licensed. One such analysis has been proposed by Hicks (2009) (see also Hornstein 2001). On Hicks’s analysis, the DP that is to become the subject is embedded in a bigger DP which also hosts an Operator, (51) (see also Kayne 2002, for a similar analysis of pronouns and their antecedents).

  1. (51)

The whole complex DP A′-moves to the edge of the embedded clause, as is expected of Operators. Via this movement step, the embedded DP is smuggled with the Operator across the embedded subject position. After this, the embedded DP, which lacks case, A-moves independently to the matrix subject position as it functions as a goal for the probe T.Footnote 30

  1. (52)

    The booksare easy (for Kate) [[Op t]to read t]

On Hicks’s analysis, TCs are thus derived via a combination of A′- and A-movement made possible by the complex structure of the embedded object DP. More specifically, two chains are formed: the complex DP forms an A′-chain with its trace position, while the matrix subject forms an A-chain with the trace inside the complex DP. Although it successfully accounts for the binding and scope properties of TCs as well as the licensing of parasitic gaps, the smuggling analysis proposed by Hicks (2009) raises a potential problem as it involves movement steps that are not generally allowed.

4.3 Improper Movement and clause size

The combined A′- and A-movement analysis proposed by Hicks (2009) could be argued to involve Improper Movement (Chomsky 1973, 1981). The ban on Improper Movement is such that if an element has moved into an A′-position, it cannot subsequently move into an A-position. Williams (1974, 2003) relates this to the functional sequence of the clause (The Williams Cycle), stating that movement into a position that is lower on the functional sequence than the position moved from is banned (see also Abels 2007; Müller 2014). The sentences in (53) are ill-formed because they involve such illegitimate movement steps: in each case, the subject has moved through a position, embedded Spec,CP, that is higher on the hierarchy than the final landing site, matrix Spec,TP.

  1. (53)
    figure bq

If TCs are derived via the movement steps proposed by Hicks (2009), it seems, prima facie, that they involve Improper Movement, as there is movement via embedded Spec,CP into matrix Spec,TP. In response to this, Hicks stresses the point that the two movement operations involve different DPs (although one is embedded under the other), creating two different chains, so that there is no Improper Movement. It is still the case though that such movement steps are otherwise ill-formed in English, as discussed by Keine and Poole (2017) (the example is originally from Abels (2007:77) who discusses it in the context of remnant movement):

  1. (54)

     *Oscarwas asked [[ how likely tto win ]it was t]

In (54), the infinitival clause (subscript 2) has undergone A′-movement to the clause edge, and the subject Oscar has A-moved out of it to the matrix subject position. In (54), the ill-formedness can be formulated as a violation of the Freezing principle, as it involves movement out of a string that has itself moved. Although the smuggling derivation proposed by Hicks (2009) differs from (54) in involving a null wh-Operator and a DP rather than a clause, the smuggling derivation proposed is sufficiently similar to the illicit one in (54) to beg the question how Freezing is circumvented in TCs.

I propose an analysis of Swedish TCs that combines A′- and A-movement but in which Freezing is not an issue and other Improper Movement effects are also avoided. On this proposal, Swedish TCs are structurally somewhere in between German and English TCs. They are like German TCs in that the object moves all the way from the embedded clause to the matrix subject position, but they are also like English TCs in that they include an A′-step in the embedded clause. According to Wurmbrand (2003), lacking A′-properties, German TCs are derived via long A-movement from the structurally reduced embedded clause into the matrix clause. In Swedish too, infinitival clauses are not full CPs, as pointed out by a reviewer and shown by Engdahl (1986:91–92). Since infinitival clauses do not have a full CP structure, embedded infinitival interrogatives are not possible, (55), and neither is pied-piped material in infinitival relative clauses, (56) (examples from Engdahl 1986:91–93)

  1. (55)
    figure br
  1. (56)
    figure bs

In TCs, the object first moves to the embedded clause edge, and then further into the matrix clause. Although the first step is to an A′-position, there is no Improper Movement, since the embedded clause edge is a position lower than a CP. I will take it to be some projection in the T domain. In that way, the movement operation does not violate the Williams Cycle. The proposed analysis accounts for the absence of intervention effects and for the fact that parasitic gaps are licensed. It also captures the scope and binding properties of TCs in Swedish.

5 Concluding remarks

This paper presents two novel findings: I show that there is a strong connection between ϕ-feature agreement and scope freezing in Swedish, and I demonstrate that Swedish TCs involve movement of the underlying object into the matrix subject position.

I follow Sauerland and Elbourne (2002) in assuming reconstructed readings are not achieved via lowering or other undoing operations, but instead to reflect the final syntactic position of an element. In Swedish, movement related to ϕ-feature agreement is syntactic but purely EPP-driven movement (not involving ϕ-feature agreement) is post-syntactic. Subjects of adjectival predicates marked for common gender or plural have undergone ϕ-agreement, while subjects of verbal, nominal and adjectival predicates marked for neuter have not. The former are interpreted in the surface position (i.e. the subject position) for scope purposes (i.e. they get wide scope relative to quantified elements lower down), while the latter can either be interpreted in their syntactic position (i.e. the object position, where they have narrow scope relative to another quantifier) or in the higher position, achieved via Quantifier Raising at LF. These patterns can be observed in TCs, passives and raising structures, all three of which come with both adjectival and verbal predicates.

The second finding concerns TCs more specifically. TCs in Swedish are like English ones in having A′-properties; they license parasitic gaps and are not sensitive to arguments intervening between the embedded object position and the matrix subject. Unlike English TCs, Swedish TCs can have a reflexive element in the subject bound in the embedded clause. For Swedish, then, there is evidence for movement of the object all the way from the embedded clause, involving an initial A′-step. Since infinitival clauses in Swedish have been shown to have a smaller structure than a full CP (Engdahl 1986), movement of the object does not violate the Williams Cycle (Williams 2003) and is therefore not an instance of Improper Movement. German TCs lack the A′-properties of English and Swedish TCs and are analyzed as being derived via long A-movement (Wurmbrand 2003). On the proposed analysis, Swedish TCs can be placed in between the English type and the German type.