Scandinavian Clause Structure and Object Shift

Peter Sells

Stanford University

Proceedings of the LFG98 Conference

University of Queensland, Brisbane

Miriam Butt and Tracy Holloway King (Editors)

1998
CSLI Publications
http://www-csli.stanford.edu/publications/

Abstract

This paper focusses on the theory of clause structure in Icelandic, contrasting it along the way with that of another Scandinavian language, Swedish. I argue that LFG provides a very simple and appealing account of two distinguishing properties of Icelandic: (i) the presence of two apparent subject positions in the `Transitive Expletive' construction, and (ii) the phenomenon of `Object Shift', where a direct object appears to the left of its expected VP-internal position.¹

There is in fact a natural correlation between (i) and (ii) in the LFG analysis, something not successfully captured in any of the many previous analyses. In Icelandic, the Transitive Expletive structure requires the IP-over-S clausal analysis that LFG provides. Object Shift, I argue, is what happens when the object is generated outside of VP. In Icelandic, this possibility arises given the availability of S and its alternative expansion as XP⁺. In contrast, Swedish lacks the Transitive Expletive construction, and so has no clause-internal S structure. Swedish Object Shift only affects weak pronouns, and this restricted Object Shift is naturally analyzed as attraction to I, again obviating the need for the pronoun to be contained within a surface VP.

The cited examples are Icelandic, except those marked with a following (S), which are Swedish.

Subject Positions

Icelandic is a Germanic language which shows VO order if both verb and object remain within the VP. Unlike many of the Germanic languages, Icelandic shows symmetric V2 structure, which means that the finite verb appears in second position in both main and embedded clauses. Examples of embedded V2 are shown in (1), taken from Holmberg (1986).

The fact that the verb (underlined) in the embedded clause is in the second position can be seen here from its position preceding the negative adverbs in these examples. The standard analysis of Icelandic developed primarily in the Government-Binding and Minimalist Program literature is that the verb appears in the highest functional head position in all finite clause types, and I will adopt this characterization, though not necessarily the details of the actual functional categories proposed in that literature.

Swedish is a `mainland Scandinavian' language, which also shows basic VO order within the VP. However, simplifying a little, Swedish only shows V2 in main clauses, as is more typical of the Germanic languages. One famous point of difference between Icelandic and Swedish is that Icelandic alone allows the Transitive Expletive construction, as seen in the examples in (2), where the verb in second position is preceded by an expletive and immediately followed by the thematic subject.

Minimalist Accounts

Bobaljik and Jonas (1996) propose that the expletive in Icelandic occupies the SpecAgrSP position, with the finite verb in the AgrS head position, and the logical subject `many Christmas trolls' in SpecTP, based on the universal grammar template postulating clause-internal categories of AgrSP, TP, AgrOP, and VP (see (3) for (2)a). Their work crucially motivates the presence of two subject positions, the specifiers of AgrS and T, with an intervening head position.

Building on Bobaljik and Jonas (1996) and other earlier work both on the Scandinavian languages and on the theory of clause structure in the Minimalist Program, Jonas (1996) and Thráinsson (1996) propose some correlating differences between Icelandic and a mainland Scandinavian language such as Swedish with regard to clause structure. In particular, they relate the position of the tensed verb and the presence of certain subject positions. These can be summarized as in (4).

With regard to (4)a, Thráinsson (1996) argues that AgrS and T are simply collapsed as I(NFL) in Swedish; Jonas (1996) makes the standard minimalist assumption that in Swedish TP lacks a specifier, and that T moves obligatorily to AgrS. As Thráinsson observes, this has the same effect as positing a single I in the first place. (4)b allows Icelandic to have the Transitive Expletive construction; naturally, in regular clauses, the subject raises through SpecTP to surface in SpecAgrSP. With regard to (4)c, we have seen that Icelandic shows verb raising in both main and embedded tensed clauses, a fact that has been taken in the Minimalist literature to show that the verb raises overtly to the highest inflectional head due to a motivation independent of that which gives V2 structures; this particular feature will not be directly relevant in the rest of this paper.

The account that I will develop below accepts (4a-c), except for notational differences. One point of contrast is that, as there clearly are two specifiers (of AgrSP and TP), standard X'-assumptions lead to the expectation that there must be two different heads which these are projected from, an expectation which is not fulfilled.

The LFG Analysis

The LFG theory of phrase structure provides functional categories as well as lexical categories. In this paper, only the functional category I(NFL) and its projection will be relevant. Functional categories differ from lexical categories in the functional annotations on their specifiers and complements. The main features of structure-function association are given in (5), from Bresnan (1997a).

For the data at hand, (c) is the crucial principle. When I takes a complement, such as VP, both I and VP are co-heads, and thus all functional information is unified--the functional domains of VP and IP are the same. Clauses where tensed verbs appear in a designated position outside of VP will have the tensed verb in I (see, for example King (1995) on Russian and Bresnan (1997a) on Welsh). The complements of the verb in I will typically appear dominated by a VP node, even though that VP has no c-structure head (for the verb is in I, by assumption).

These properties can be illustrated for Swedish, in which the structure of the clause is such that the complement of IP would be VP. This analysis predicts all of the Swedish properties in (4), providing single head and specifier positions above VP. The schematic clause structure is shown in (6)a, with the structure of a simple example Anna såg boken (`Anna saw the book.') in (6)b.

(5)b illustrates the V2 effect in main clauses: if there is a single tensed verb, it will appear in I. LFG allows projective phrases such as VP to lack a c-structure head, but requires that all projective phrases have an `extended head' (see Bresnan (1997a)). I is the extended head of VP in (5)b as it is a c-structure head and IP maps to the same functional domain as VP.

In the schematic structure in (5)a, the position V⁺ will be filled by a non-finite verb just in case there is an auxiliary verb in I. By the principles in (5), there is only one SUBJ position in (5)a, namely SpecIP. SUBJ cannot be associated with a phrasal position internal to VP, and so it is the non-SUBJ functions that will be generated as XP daughters of VP.

Bresnan (1997a) identifies two forces in syntax--`endocentricity' and `lexocentricity'. The former provides headed structures according to X'-theory, such as the projections for IP and VP just considered. The latter "associates syntactic functions directly with features borne by words rather than with the configurational relations of phrases in syntax" (p.103). The syntax here employs a non-projective category, S, a category lacking a c-structure head, which hosts words and phrases in `lexocentric' environments. In radically non-configurational languages, S expands as C⁺, that is, some number of categories of any type and bar level. S may also expand as NP - XP, giving a canonical and configurational subject-predicate structure.

Combined with the endocentric structures, the theory offers a range of typological possibilities as to the surface manifestation of grammatical functions: at least through structural position, relative order, head-marking, and dependent-marking. While none of these options are mutually exclusive, general principles of economy of expression and structure-function mapping lead us to expect languages to effectively choose some properties and not others. For example, radically non-configurational languages probably do not manifest the NP - XP expansion for S, while the Celtic languages show only that expansion for S.

S can be the complement to I, and when it is, it is a co-head with I, by (5)c. This is structure by which LFG provides `internal subject' structures, where S expands as NP - VP. Given such a possibility, the properties of Icelandic subject positions would be analyzed perfectly on the assumption that Icelandic has an IP structure with an S complement. This structure will provide two `subject positions', SpecIP and the subject of S, with an intervening head position, I, where tensed verbs will be generated. However, as S can have no X-0 head, there will be no expectation of a second head position (corresponding to the head of TP in (3)). We will see in all the examples below that S dominates only phrasal (XP) constituents. The schematic structure of the clause is shown in (7); again the c-structure head of VP, V⁺, is filled only by a non-finite verb in the presence of a tensed auxiliary verb in I.

With these assumptions, (8) is the structure of (2)a.²

In the case of a prototypical clause with a subject only in SpecIP, usual principles of structural economy will favor a simple VP complement to I, over a structure with S dominating only a VP, giving a structure just like (6)a.

At this level of analysis, the IP-S structure in LFG perfectly models the AgrSP-TP structure proposed in the Minimalist literature, with the exception of not making the prediction that the head positions of AgrS and T ought to be filled independently of each other. The upper parts of (3) and (8) are essentially identical, though the structures diverge radically below T' and VP respectively. The following section takes up the question of this lower structure, examining some other properties of the non-projective category S.

Flat vs. Hierarchical Structures and Object Shift

Kaplan and Zaenen (1989)

An important precursor to the analysis I present below can be found in Zaenen and Kaplan (1989), who independently proposed an S structure for Icelandic. They noticed that Icelandic shows evidence for a flat S structure when there is a single tensed verb in the clause, but when there is a tensed auxiliary, the non-finite verb heads a structured VP.

The facts that motivated their analysis are as follows. An adverb such as sjaldan (`seldom') may flank a VP, but not appear between constituents of VP, when there is an auxiliary. This is seen in the contrast in the examples in (9), with the adverb underlined.

However, when the main verb is in second position, not in VP, the adverb can appear between any constituents:

This shows that there is an analysis of (10) that does not appeal to a surface VP. Updating the structures of Zaenen and Kaplan (1989) to fit with current LFG, their structure for (10) would be as in (11).

The NP and PP dominated by S both fulfill argument functions here, as the non-subject complements of the verb. This expansion of S is different from the predicating NP - VP expansion discussed above; (10) shows that Icelandic must allow a expansion of S where each daughter phrase has an argument or adjunct function.

However, this rather simple conclusion is challenged by Holmberg (1985), who observes that the data in (10) could be handled by an account with a structured VP in deep structure, with transformations of verb movement to INFL and an Object Shift rule applying only to NPs. A base-generated version of This account would give a structure just like (11) except that the PP would be the sole constituent within a VP, where that VP is sister to the NP `the butter'. Although I cannot do justice to the rationale for this slightly more complex analysis here, I will adopt it below in section 2.4 and subsequent sections.

The Phenomenon of Object Shift

Object Shift is a phenomenon which shifts direct objects leftwards out of the VP, where the leftward movement is visible due to the position of negation and other adverbs. The examples in (12)-(13) from Holmberg (1986, 217) illustrate Object Shift in Icelandic within embedded clauses, where the verb nevertheless raises into the V2 position. Object Shift is generally optional in Icelandic, so the pairs of examples show the unshifted and shifted versions, with the object underlined.

In Icelandic, all types of direct object NPs can undergo Object Shift. In Swedish, only pronominal objects undergo Object Shift.³ Object Shift has been associated with a correlation known as `Holmberg's Generalization', following the pioneering work of Holmberg (1986), and it says that Object Shift is possible just in case the V has raised out of the VP to the highest functional head position.

The following Swedish examples from Holmberg (1997) illustrate pronominal Object Shift; if the main verb does not leave VP, due to the presence of an auxiliary, as in (14)b, or due to the fact that there is no verb movement in an embedded clause at all, as in (14)c, Object Shift is not licensed. Following Holmberg's representations, I indicate the base positions of the verb and object by traces, though such empty elements will not feature in the analysis that I propose.

According to Holmberg (1997), Object Shift is somewhat optional in (14)a, so the corresponding unshifted version in (15)a is acceptable to some speakers; for all speakers the acceptable versions of the (b-c) examples involve no Object Shift.

As just noted, Object Shift applies in principle to all NPs in Icelandic, and is somewhat optional;⁴ in Swedish, it applies only to pronominal objects, and is largely obligatory. The variable optional/obligatory nature of Object Shift seems to be related to aspects of semantic and pragmatic interpretation depending on whether the object is in the VP or not. I will ignore these interpretive effects in this paper; for discussion, see Diesing (1997) and Vikner (1997).

In the examples so far, the shifted object immediately follows the tensed verb, giving the appearance that the object has moved to a designated `Object Shift position'. This intuition has guided many of the recent transformational analyses. However, there is no evidence for such a distinguished position, and in Icelandic, at least two XP constituents may intervene between the inflected verb and the shifted object. First, it is possible to have both the Transitive Expletive structure and Object Shift (Bobaljik and Jonas (1996, 213-4)). As shown below the first example, the thematic subject in the Transitive Expletive structure follows the verb in I, but precedes the shifted object. This relative positioning is one of the main factual points that Bobaljik and Jonas (1996) wished to establish.

Second, with double-object verbs, the indirect object must precede the direct object, regardless of whether the direct object is shifted or not. With both objects in situ, the indirect object (underlined in (17)) must immediately follow the V head of VP:

The indirect object may appear in front of negation, with the direct object apparently still in VP, following negation, as shown in (18)a. However, reversing the objects leads to ungrammaticality, as in (18)b.

Additionally, subject to various pragmatic felicities associated with intonational patterns (see Collins and Thráinsson (1997)), both objects may shift. In these cases, the indirect object must precede the direct object; this is the same order that is found when both objects remain within the VP.

In fact what these data show is that the indirect object and direct object are always at the same level: whether both within the VP, or both out of the VP, the indirect object must precede the direct object, as (19) confirms. Overall, (17)-(19) indicate that when both objects shift, the indirect object must precede both negation and the direct object, but the relative order of the latter two constituents is somewhat free. On the correct assumption that negation cannot appear within the VP, naturally negation will precede both indirect and direct objects when they remain within the VP, as in (17).

Summing up, we have the following generalizations about constituent order:

Additional Facts

Holmberg (1997) presents some important new facts concerning Object Shift, illustrating with examples of pronominal object shift from Swedish. Crucially, verb raising is shown to be a necessary but not sufficient condition on Object Shift; what Holmberg shows is that the correct generalization is that there should be no overt material whatsoever within the VP.

For example, a particle left within the VP blocks Object Shift.⁵

(21)a shows that V movement is not a sufficient condition on Object Shift: rather, there must be nothing overtly within VP (or, to be more precise, there should be no overt material between the left edge of VP and the base position of the object, within VP, according to Holmberg).

Additionally, and strikingly, Holmberg (1997) shows that V-topicalization in the presence of an auxiliary can license Object Shift:

This example is important, as it shows that V-raising through the functional heads of the clause is irrelevant to the licensing of Object Shift; all that matters is that there is no verb in VP. Fronting of other elements also provides the conditions for Object Shift to occur:

Holmberg's proposal for the analysis of Object Shift that these new facts necessitate is one that involves not a syntactic movement operation, but rather `PF-movement'--movement of just a phonological matrix. This movement is subject to the principle of Relativized Minimality (see Rizzi (1990)), extended to allow for what Holmberg presents as a `Relativized Minimality effect pertaining to PF-operations'. This has the effect that `PF-operations cannot move a phonological matrix over another phonological matrix'. Along with some other assumptions (see Holmberg (1997, 214)), this blocks Object Shift when there is intervening overt phonological material in the VP. The motivation for the movement is that pronouns are specified as non-focal ([-Foc]), and must move out of the focal domain. In Holmberg's proposal, all non-focal elements move at least covertly out of the focal domain, and pronouns move overtly.

This proposal may be descriptively adequate, but it expains little about the licensing conditions for Object Shift, as it involves essentially new theoretical devices just for the purpose of Object Shift. Holmberg suggests that some properties of the proposal would relate to conditions on other PF-operations, such as cliticization or contraction. This has some credence for Swedish, where the elements that undergo Object Shift are weak pronouns, which levitate towards I, but it fails completely to offer any insight for Icelandic, where full NPs undergo shifting. Holmberg suggests that the category-internal feature structure of all Icelandic NPs is like that of Swedish pronominals. Even if this were true--and could be shown to be true--it fails to relate Object Shift in Icelandic to the availability of the Transitive Expletive construction.

The LFG analysis

Prefacing the analysis to come, the basic idea is extremely simple: the object `shifts' when there is no VP to contain it, as would be possible just in case there were no other overt elements that required a VP to contain them. This is possible in Icelandic due to the availability of S, shown by the adverb facts from Zaenen and Kaplan (1989) and the Transitive Expletive construction. In Swedish, there is no possibility for S, but VP need not be present to host an object pronoun, in certain restricted circumstances, described below. This simple idea does not provide a fully general account, though, and a slightly more articulated structure can be shown to be present in some cases of Object Shift.

The unifying generalization about Object Shift brought out in Holmberg (1997) is that Object Shift is not possible if any overt element which should precede the object is present in VP. This generalization might follow naturally in the LFG account--VP would be absent if it dominated no overt material, as there are no traces in c-structure, and natural principles of structural economy will not allow a VP to be projected if it is not necessary (cf. Bresnan (1997a) `Economy of Expression'). However, the facts of PP and Particle positioning (section 2.6) show that there may be a surface VP even if an object has `shifted' out of it, so long as what `remains' in VP would have followed the object in the unshifted structure.

To make the analysis more specific, I propose that there are 3 possible expansions for S, labelling the new option the `multi-NP' option. The multi-NP option is a kind of `semi-configurational' structure, halfway between the fully non-configurational S introduced in Kroeger's (1993) analysis of Tagalog and the regular predicating or small clause structure:

In the multi-NP expansion, any functional annotation on NP is possible with the exception of the pure discourse functions of TOP and FOC. Hence, any NP argument that could appear in a projective endocentric structure can also appear in the multi-NP S. Clearly Icelandic is not a non-configurational language, so (24)a is not available; that expansion is primarily motivated for `radically non-configurational languages' as are found in Australia (see e.g. Nordlinger (1998)). The other two expansions of S are available--the predicating option for the Transitive Expletive structure (without Object Shift), and the multi-NP option for Object Shift. Finally, as there are alternate expansions for S, and the possibility of VP complement to I, Object Shift is not obligatory; even with the verb in I, nothing prevents a structure in Icelandic with a VP containing the object (in the simplest case, parallel to (6)b).

To reiterate, the leading idea of the LFG analysis is that Object Shift will be possible just in case a clausal structure with an object outside of VP can be generated. This is what unifies all the Scandinavian languages; the extra feature of Icelandic, that it has S, gives it more structural possibilities than the other languages. In Swedish, Object Shift only occurs with weak pronouns: plausibly these weak pronouns can appear dominated by I in the c-structure, as can negation.

The crucial question, of course, is where the `multi-NP' expansion comes from, and how to motivate the restriction to only NPs. Following a suggestion by Joan Bresnan, I believe that this motivation can be thought of along the following lines, though the ideas are not well-developed yet. Taking the idea of constructive case as developed in Nordlinger (1998), let us suppose that the rich case marking on Icelandic NPs allows their GF to be determined without reference to a specific structural position--any easy supposition, given the relative freedom of NP placement that we observe. Then, the predicating expansion of S would be what we find in a language like English which has only configurational determination of GFs, while further options for daughters under S would be available to due to rich case marking in Icelandic. (Another way to think about this would be that the multi-NP option is available in English, but as the subject must be defined as `the sister of VP', no more than one instantiation of NP is possible.)

Icelandic

As we have seen, Icelandic allows S as a complement to IP. If S expands as NP - VP, with VP annotated as a co-head, we will have a standard structure with a fully articulated VP. Negation and other adverbs will be left-adjoined to that VP. However, another option for S is to expand as NP⁺ - VP, where each NP is annotated with an argument function. Such a structure is shown schematically in (25).

The next step is to get the order of constituents correct. Some of the generalizations about this order were given in (20) above. Strictly speaking, the analysis that I offer here requires the ordering principles to be ranked in terms of their violability, as would naturally be expressed as Optimality Theory constraints, but I will not formalize that part of the analysis in this short paper.

The standard assumption for ordering principles is that they only hold between sister elements. However, there are two structures in LFG over which contraints can be stated: c-structure or f-structure. I will propose that principles stated in terms of c-structure categories hold among mutually c-commanding elements, while those in f-structure hold among mutually f-commanding elements.⁶

I will interpret (a-d) as ranked: looking at the functional constraints, the TOPIC (in SpecIP) will be leftmost in the whole clause, and the SUBJ will always precede an IOBJ. As I am simplifying here largely to just the order of NPs, IOBJ will precede OBJ, due to (26)d. For structures with an an X-0 c-structure head, that head will always be first in the sequence due to (26)b.

Specifically, within VP, these principles give the order V - IO - DO - X. Negation, which I take here to be left-adjoined to VP, will of course precede all daughters of VP. Within IP, the specifier will be initial by (26)a, and within I', the head will be initial by (26)b. For S, we have two options: the predicating structure, which (26)c will order as NP - VP, for NP carries the SUBJ information. In the multi-NP structure, the real effects of (26)b-c are seen. The facts to be accounted for are schematized in (27).

These are the only grammatical orders with at least one object under S. The order SUBJ - Neg/Adverb - OBJ is grammatical, but is generated with the predicating expansion of S, within which OBJ is the sole daughter of a co-head VP. Also of interest is (17), which has the order V - IOBJ - OBJ - X. This shows that the same ordering principles (specifically (26)d) apply to both `unshifted' and `shifted' structures, and that the relative order of IOBJ and OBJ is preserved regardless of their c-structure relationships, as the functional ordering constraints apply to mutually f-commanding elements, namely any elements within the same nucleus. The OBJ can only precede the IOBJ if the OBJ is also TOPIC, in SpecIP, as predicted by the fact that (26)a outranks (26)d.

This approach contrasts with the movement-based approach in Müller (1998), which has a constraint to the effect that the base ordering relationships between elements are preserved at later derivation levels: the effects of this constraint follow directly on a base-generated account, where different structural configurations are subject to exactly the same ordering principles (e.g., those in (26)).

Swedish

With regard to the placement of the shifted pronominal in Swedish, the key facts are shown in (28).

What we see here is that the weak pronouns are attracted to I. They appear with the verb in I just in case the verb has not moved up to C, as in (28)a; alternatively, the verb is in C, followed by the subject and then the shifted pronoun in I, as in (28)b. Accepting this much, along with the common observation that negation is attracted to I (for an account using LFG and Optimality Theory, see Bresnan (1997b)), the gross structures for these examples will be as in (29).

If the pronoun and negation can be generated under I, then there need be no VP, and we have the Object Shift structure. It is important to note that the pronouns and negation do not directly join up with the verb, for the verb alone appears in C, as in (29)b which represents (28)b. This separation shows that pronominal Object Shift is not true incorporation into V, as discussed in Vikner (1994, 504) and Kaiser (1997, 104-5). As an adjunction structure, it is reasonable that only X-0 elements can adjoin to I, with full NPs generated only in phrasal argument positions.

How can the pronominals in (29) be associated with a grammatical function? Sadler (1997) suggests that the natural principle of association is that a pronominal adjoined to an X-0 can be associated with whatever GFs would be normally expressed within the phrasal projection of the head. I expand on her ideas, and propose (30), where F refers to a functional category, using the notion of extended head (see Bresnan (1997a)). Here, only the functional category I is of relevance.

In (29)a, the head of IP is V; in (29)b the V in C is the extended head of the IP. In both cases, the pronominals in (29) will be associated with any non-SUBJ argument function, the argument functions which are expressed within VP.⁷

Now we need to find an account of the examples in which particles block Object Shift. I can only sketch the analysis here, and will do so with respect to (21)a.

However it is formulated, there must be a Swedish-specific principle of ordering that requires the particle to precede the pronoun, as mentioned in section 2.3. I make the natural assumption that the particle cannot be generated adjoined to I, so there must be a VP to host it. Intuitively, the result we want is that if there is a VP, the particle should precede the pronominal under that VP; hence (21)a is out. However, we want to allow a pronominal adjoined to I to appear in the same clause as material dominated by VP which should follow the pronominal, such as the full (oblique) PP in (31).

The difference is that the particle in (21)a is an X-0, and plausibly a co-head; clearly the PP in (31) is not. Taking the idea of `lexical expression' from Bresnan (1998), we can develop a notion of a `functional head' and its relative ordering which will correctly discriminate between the two relevant structures. What we are interested in are those X-0 categories--lexical expressions--which correspond to the functional head of the clause. Formally, these will be X-0 elements for which a path of head annotations can be traced to the highest projection of the clause. Let us call this the `f-head':

In (21)a, the f-head is { I, Prt }; the verb is in I, and the Prt is the sole constituent of VP. In (31), the f-head is { I }. What we see from both examples is that the every member of the f-head must precede the complement pronoun. As we are using a system of ranked constraints, we can state an f-head ordering constraint to this effect, stating it as generally as possible, and let higher ranking constraints interact with this one, as in (33).

The upshot of this account of the pronouns is this: a pronoun can be generated adjoined to I (giving object shift), or it can be generated within the VP. There are preferences concerning these two positions which I have not addressed here, and which seem to be subject to dialectal variation.⁸ Although (33) is `minimally' simple, it has exactly the desired effect: a pronoun (in fact, any category) must be within the VP--hence unshifted--if there are other elements within VP which are f-heads which are in VP, such as a main verb or a particle. This captures neatly `Holmberg's Generalization' and accounts for the contrast between (21)a and (31)).

Conclusion

In this paper I have shown that the LFG IP-S structure provides a convincing account of the availability of the Transitive Expletive and Object Shift constructions in Icelandic. The analysis of Object Shift crucially relies on the ability of objects to appear outside of VP, a possibility in a language which has S. In constrast to Icelandic, the fact that Swedish allows Object Shift only for pronominal objects follows from the natural assumption that such objects can be associated with I, again obviating the need for them to be within VP. A full account of Object Shift will require a more complete account of ordering principles and the domains of their application, and I have offered some suggestions as to what those might be. Although I have not provided any explicit discussion, this analysis of Object Shift compares favorably to recent proposals in the Minimalist Program, and provides a strong argument for the multi-NP expansion of S in LFG.

Bibliography

Bobaljik and Jonas 1996

Bobaljik, Jonathan, and Dianne Jonas. 1996. Subject positions and the role of TP. Linguistic Inquiry 27, 195-236.

Bresnan 1997a

Bresnan, Joan. 1997a. Lexical-functional syntax: Inflectional morphology and phrase structure variation. Chapter 5 of forthcoming book. Ms. Stanford University.

Bresnan 1997b

Bresnan, Joan. 1997b. Optimal syntax. Ms. Stanford University. To appear in The Pointing Finger: Conceptual Studies in Optimality Theory, download from http://www.leidenuniv.nl/hil/ot/.

Bresnan 1998

Bresnan, Joan. 1998. Morphology competes with syntax: Explaining typological variation in weak crossover effects. In Pilar Barbosa et al. (ed.), Is the Best Good Enough? Optimality and Competition in Syntax, 59-92. Cambridge, MIT Press.

Collins and Thráinsson 1996

Collins, Chris, and Höskuldur Thráinsson. 1996. VP-internal structure and object shift in Icelandic. Linguistic Inquiry 27, 391-444.

Diesing 1997

Diesing, Molly. 1997. Yiddish VP order and the typology of object shift in Germanic. Natural Language and Linguistic Theory 15, 369-427.

Holmberg 1985

Holmberg, Anders. 1985. Icelandic word order and binary branching. Nordic Journal of Linguistics 8, 161-195.

Holmberg 1986

Holmberg, Anders. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Stockholm, University of Stockholm, Department of Linguistics.

Holmberg 1997

Holmberg, Anders. 1997. The true nature of Holmberg's generalization. In Kiyomi Kusumoto (ed.), Proceedings of NELS 27, 203-217. Amherst, GLSA.

Jonas 1996

Jonas, Diane. 1996. Clause structure, expletives and verb movement. In W. Abraham, S. Epstein, H. Thráinsson, and C. J.-W. Zwart (eds.), Minimal Ideas, 167-188. Amsterdam, John Benjamins.

Josefsson 1992

Josefsson, Gunlög. 1992. Object shift and weak pronominals in Swedish. Working Papers in Scandinavian Syntax 49, 59-92.

Kaiser 1997

Kaiser, Lizanne. 1997. The morphological cliticization of object-shifted weak pronouns in Swedish. In Lizanne Kaiser (ed.), Yale A-Morphous Linguistics Essays, 99-129. New Haven, Department of Linguistics, Yale University.

Kaplan and Zaenen 1989

Kaplan, Ronald M., and Annie Zaenen. 1989. Long-distance dependencies, constituent structure, and functional uncertainty. In Mark Baltin and Anthony Kroch (eds.), Alternative Conceptions of Phrase Structure, 17-42. Chicago, University of Chicago Press.

King 1995

King, Tracy H. 1995. Configuring Topic and Focus in Russian. Stanford, Dissertations in Linguistics, CSLI Publications.

Kroeger 1993

Kroeger, Paul. 1993. Phrase Structure and Grammatical Relations in Tagalog. Stanford, Dissertations in Linguistics, CSLI Publications.

Müller 1998

Müller, Gereon. 1998. Order preservation, parallel movement, and the emergence of the unmarked. Ms. Universität Stuttgart (available via the Rutgers Optimality Archive at http://ruccs.rutgers.edu/roa).

Nordlinger 1998

Nordlinger, Rachel. 1998. Constructive Case: Evidence from Australian Languages. Stanford, Dissertations in Linguistics, CSLI Publications.

Rizzi 1990

Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, MIT Press.

Sadler 1997

Sadler, Louisa. 1997. Clitics and the structure-function mapping. Ms. University of Essex, download from http://clwww.essex.ac.uk/louisa/.

Thráinsson 1996

Thráinsson, Höskuldur. 1996. On the (non-)universality of functional categories. In W. Abraham, S. Epstein, H. Thráinsson, and C. J.-W. Zwart (eds.), Minimal Ideas, 253-281. Amsterdam, John Benjamins.

Vikner 1994

Vikner, Sten. 1994. Scandinavian object shift and West Germanic scrambling. In Norbert Corver and Henk van Riemsdijk (eds.), Studies on Scrambling: Movement and Non-Movement Approaches to Free Word-Order Phenomena, 487-517. Berlin and New York, Mouton de Gruyter.

Vikner 1997

Vikner, Sten. 1997. The interpretation of object shift, optimality theory, and minimalism. Working Papers in Scandinavian Syntax 60, 1-24.

Footnotes

 

(1)

I am grateful to Avery Andrews, Joan Bresnan, Lizanne Kaiser and Yukiko Morimoto for useful comments and suggestions. This short paper represents an intermediate stage of my thinking on the phenomena and issues here--a much longer work is in preparation.  

(2)

Though not shown here, I assume that the expletive lacks an independent PRED and maps with the thematic subject to the SUBJ function.  

(3)

Norwegian, Swedish, and Danish all show pronominal Object Shift.  

(4)

All object NPs in Icelandic can undergo Object Shift, regardless of their surface case (see Collins and Thráinsson (1997, 399)).  

(5)

In the other Scandinavian languages, a particle cannot appear between a verb and a pronominal object. Swedish differs in that only (21)b is acceptable (see Josefsson (1992, 87)). This fact will be important in section 2.6.  

(6)

Strictly speaking, we need to appeal to `f-precedence' in the latter case.  

(7)

Both indirect and direct object pronominals can adjoin to I.  

(8)

See Kaiser (1997, 115-125) for a thorough discussion of these differences.