2. OBJECT SHIFT

The term object shift is commonly used to refer to the placement of an object pronoun to the left of an adverb, such as negation, illustrated in (2a–a'). Only if the lexical verb has raised out of VP, as in (2b), is object shift possible (Holmberg, Reference Holmberg1986, Reference Holmberg1999; Vikner, Reference Vikner, Everaert and van Riemsdijk2005).

1. (2)

This interplay of verb movement and object shift is known as ‘Holmberg's Generalization’.

3. LONG OBJECT SHIFT

The term long object shift (LOS) is commonly used to refer to the placement of an object pronoun to the left of the subject, as in (3) (examples in (3a, b) are from Holmberg Reference Holmberg1986).

1. (3)

It is clear from the examples in (3) that 1st and 2nd person pronouns can undergo LOS. The next section takes a closer look at reflexives and LOS.

4. LOS AND REFLEXIVES

Holmberg (Reference Holmberg1986), Josefsson (Reference Josefsson1992) and Teleman, Hellberg & Andersson (Reference Teleman, Hellberg and Andersson1999) all claim that weak (unstressed) reflexives can undergo long object shift, as in (4a–b). The reflexive pronoun can even precede, i.e. raise above a (focused) subject pronoun, as in (4c).Footnote ¹ This does not hold for all reflexives, as is evident in (5).

1. (4)

2. (5)

The conclusion we can draw from (5) is that long object shift is not allowed with 1st and 2nd person reflexives. Recall from (3) that it is possible with 1st and 2nd person object pronouns. Note that Swedish object pronouns and reflexive pronouns have the same forms in 1st and 2nd person. This similarity in form has led some people to treat all these items as members of the same pronoun category. I will return to this similarity in section 10.

5. QUIRKY LOS

We now return to the mismatch in agreement that we saw in sentence (1). The following sentences show that long object shift of reflexives is possible with 1st and 2nd person antecedents, but the funny thing is that the reflexive has to be 3rd person. I use the term quirky LOS to refer to constructions involving long object shift and no person-agreement between the reflexive and its antecedent.Footnote ²

1. (6)

There are more requirements on the LOS of 1st and 2nd person reflexives. The next section deals with the conditions that must hold on the antecedent in this construction.

6. PROPERTIES OF QUIRKY LOS

7. A SYNTACTIC ANALYSIS

Cardinaletti & Starke (Reference Cardinaletti, Starke and van Riemsdijk1999) make a distinction between three types of pronouns: strong pronouns, weak pronouns, and clitics, illustrated in (12).Footnote ⁵ Their basic claim is that the more structure a pronoun is missing, the further up it must move to recover its structure. Clitics lack the most structure and will, as a consequence, move furthest.

1. (12)

   1. a. Strong pronouns

      

   2. b. Weak pronouns

      

   3. c. Clitic pronouns

      

The structure of weak elements is thus a ‘peeled’ structure of the next higher strong element. The reason a deficient element is chosen over a more structurally complex one is, according to Cardinaletti & Starke (1999:198), an ‘economy’ condition:

A possible analysis of quirky LOS is to assume that there is only one Swedish reflexive clitic, sej.Footnote ⁶ There is agreement between the antecedent and the reflexive at some point during the syntactic derivation, (13a), satisfying binding principle A (see e.g. Reuland Reference Reuland2001; Heinat Reference Heinat2006 for analyses of binding involving Agree) but at lexical insertion there is only one clitic form available for the highest – i.e. the long object shifted – position.

1. (13)

   1. a. subject [V P verb refl]

   2. b. [CP verb refl subject . . .]

Thus, when we have the order in (13b), i.e. a long object shifted reflexive, sej is the only possible form to insert. However, this analysis fails to account for the fact that in order for the reflexive to long object shift across a 1st or 2nd person pronoun, the antecedent has to be modified, as we saw in section 6.1. Consequently, the prediction of this analysis is that sentences such as (14) are well formed.Footnote ⁷

1. (14)

8. A CONCEPTUAL SEMANTIC ANALYSIS

Since the antecedent must be modified in a way that implies a discourse context of 3rd person, it is tempting to try to account for the data in a semantically-oriented framework. In conceptual semantics (such as Jackendoff Reference Jackendoff1990), given a sentence like (15), the antecedent has approximately the conceptual structure as in (16) (notation from Culicover & Jackendoff Reference Culicover and Jackendoff2005).

1. (15)

2. (16) [CUT ([not only 1st, but also [OTHERS]], REFL)]

Either, or both, of the parts of the subject arguments can appear in syntax (Culicover & Jackendoff Reference Culicover and Jackendoff2005:381). However, assuming that [OTHER] is visible in the syntax would predict that sentence (17) is well formed, since the reflexive should be bound by OTHERS.

1. (17)

In (17), the reflexive hasn't shifted but since the order between the reflexive and its antecedent is irrelevant in a semantic analysis, in contrast to a syntactic analysis, there is nothing that can rule out sentences with a mismatch in agreement with the reflexive in its base position.Footnote ⁸ The conclusion is that even though the implied 3rd person can be accounted for in this type of framework, a (conceptual) semantic analysis makes no distinction between the order of the antecedent and the reflexive, and this order has to be stipulated.

9. A MORPHOLOGICAL ANALYSIS

What seems to be needed in an account of the data involving the reflexive and LOS are morphological processes that take into consideration both the linear order of the reflexive and the antecedent, and the mismatch between their ϕ-features. In her analysis of Romance clitic clusters, Bonet (Reference Bonet1995) makes use of exactly these kinds of processes. She wants to account for, among other things, the unexpected form se in (18a), from Spanish. The expected form is the dative clitic le we see in (18b).

1. (18)

Bonet (Reference Bonet1995) makes use of the morphological processes feature delinking (which may result in deletion) and feature insertion. These processes take place when certain types of clitics cluster. She also claims that ‘[p]ronominal clitics constitute hierarchical structures of unordered morphological features’ (p. 614). In line with Bonet's assumptions about the structure of Romance clitics, I assume that the structure of Swedish reflexive pronouns/clitics is the following:Footnote ⁹

1. (19) Reflexive

   

When the syntactic component gives the post-syntactic morphophonological component the order reflexive – antecedent, i.e. when the reflexive has undergone long object shift, as in (20), the following processes takes place: the 1st or 2nd person feature is first delinked and later deleted by stray erasure, (21) (from Bonet Reference Bonet1995:633). This is only possible if the antecedent has a 3rd person feature, implied or explicit, and there will be no delinking if the antecedent has 1st or 2nd person feature. Any sequence of a reflexive with 1st and 2nd person features and an antecedent with those features will be ruled out by morphological rules, in line with Bonet's analysis.

1. (20)

2. (21)

Bonet claims that in addition to deleting features, it is possible to insert features into the clitic structure (Bonet Reference Bonet1995:631–633). What we find in Swedish, in (22), is that a 3rd person feature is inserted, just as in Bonet's account of Italian, where she claims that a 1st person feature is inserted into clusters such as the one in (23).

1. (22)

2. (23)

In (23c), a feature 1st is inserted and since impersonals are inherently specified for plural (p.631) the resulting structure is the same as 1st person plural, i.e. ci ‘we’. In contrast to Bonet's analysis of Italian, we have a reason for why in Swedish a 3rd person feature is to be inserted into the structure in (22). This feature is only licensed in a context where a third person is implied. Since this feature is inserted in the morphophonological component, it has no effect on agreement relations in the syntax. Also, the fact that the insertion of features has to be licensed somehow (contra Bonet) accounts for the fact that sej can never precede antecedents that fail to imply a 3rd person.Footnote ¹⁰

10. CONSEQUENCES

One conclusion we can draw from the feature mismatch between antecedent and reflexives that we have seen in this paper is that there is a difference between 1st and 2nd person pronouns and 1st and 2nd person reflexives. If these two types of pronoun were the same, as, for example, Reuland (Reference Reuland2001:464–465) claims, we would expect this mismatch in agreement also in sentences such as (24), exemplified in (25), where a personal pronoun has long object shifted and is adjacent to a DP that not only implies but also syntactically has a 3rd person feature.

1. (24)

2. (25)

In (25), we have the following linear order:

1. (26)

Delinking of the 1st or 2nd person feature and insertion of the 3rd person feature are not allowed, and this indicates that there is a difference between 1st and 2nd person pronouns and reflexives. Only reflexives can have their person features delinked. Exactly what this difference is is not the topic of this paper, but see Heinat (Reference Heinat2006) for an analysis where the difference between reflexive and personal pronouns lies in the internal syntactic structure of the two types of pronoun.

Another conclusion is that the quirky long object shifted sej in Swedish supports Nevins (Reference Nevins2007) claims that some morphophonological processes cannot be accounted for without making reference to a 3rd person feature, even if it seems to be possible to do so in the syntax. In the case of Swedish it is obvious that the delinking and deletion of the reflexive's person features can only happen in the context of a 3rd person feature even if that feature is not visible to agreement effects in the syntactic derivation. Also the 3rd person feature that is inserted in the structure of the reflexive cannot be visible in the syntax. Consequently, both the delinking and the insertion make reference to a third person and they both take place as post-syntactic morphological processes.