Head Conjuncts: Evidence from Old Swedish

This article deals with the size of coordinate conjuncts. More specifically, it aims to answer the following question: can conjuncts be smaller than phrases—that is, heads? In the previous literature, it has been maintained that head conjuncts indeed exist (Borsley 2005, te Velde 2006, Zhang 2010). If this were correct, any theory of coordination that presupposes that conjuncts are always phrases (see, e.g., Kayne 1994, and similar approaches as in Johannessen 1998) would have to be rejected.

However, I will show that unambiguous head conjuncts are harder to come by than appears to be the case at first. What looks like head coordination could in fact be coordination of phrases that have been vacated of everything except the head prior to being moved to the left (remnant movement). In order to find truly nonphrasal conjuncts, I turn to the Scandinavian construction known as stylistic fronting (SF). In a subtype of SF that occurs in Old Swedish, the fronting takes place in the presence of an overt (albeit pronominal) subject. In this particular context, no unambiguous phrases (i.e., units larger than a single word) are ever fronted, only single words; see (1). Here, two such elements, the nonfinite verb ofra ‘sacrifice’ (in (1a)) and the simplex object þianist ‘service’ (in (1b)), have moved from their base position in the V domain to the left of the finite verb. Remnant movement of a phrase (as suggested by Franco (2009) and Ott (2009) for SF in general) cannot be justified here, since nothing that is clearly a phrase is ever targeted. Head movement is thus the only possible analysis.

I will argue that the pronominal subject cliticizes to the complementizer (following Platzack 1988), thereby not disturbing the required adjacency between the finite elements (i.e., the complementizer and the finite verb) and the fronted element. This adjacency requirement (originally formulated by Bošković (2001) and modified slightly here) is, in a way, as puzzling as a purely descriptive empty-subject requirement; nevertheless, the adjacency account makes correct predictions about the distribution of SF that reference to empty subjects cannot.

Sometimes, the fronted heads in clauses with cliticized subjects originate in a coordinate structure; this is shown in (2), where the first of two conjoined nonfinite verbs (which share the sentence-final objects) is stylistically fronted.

I will argue that SF is an operation that takes place after Spell-Out, in PF (following Sigurðsson (2010) and also drawing on ideas put forward by Egerland (2013)). Consequently, although the coordinate split in (2) appears to violate the Coordinate Structure Constraint (CSC; first formulated by Ross (1967)), because the entire first conjunct moves to the left, stranding the rest of the coordinate structure, narrow syntax, where the CSC applies, is not involved in the actual fronting of the first conjunct. Still, what leads up to the fronting is certainly syntactically relevant. In order to feed the PF component proper material for SF in clauses with a pronominal subject, the syntactic component needs to be able to derive a coordinate structure in which the first conjunct is a head. If it were only able to derive phrasal conjuncts, (2) would not be possible.

Certainly, the fronted head in (2) could be part of a phrasal conjunct, out of which it has been extracted; indeed, unilateral extraction from the first conjunct, usually referred to as a violation of the across-the-board restriction (first described by Ross (1967)), is attested in other contexts in older Swedish—for instance, relativization (Magnusson (Petzell) 2007, Petzell 2010). However, such an analysis predicts that SF of conjunct-internal material should occur in general, which is not the case. Also (and crucially), a phrasal first conjunct in (2) would contain a gap corresponding to the shared object. Since such gaps need to be prosodically licensed locally by the governing verb (Hartmann 2000, te Velde 2006), the fronting of saalth ‘sold.PTC’ (even if it does occur in PF) is predicted to be impossible, contrary to fact.

The article is organized as follows. I begin in section 3 with an overview of various coordinate data that appear to indicate that a conjunct can be smaller than a phrase. I argue, however, that while nonphrasal status is certainly feasible in many cases, unambiguous examples are hard to come by. In section 4, I address SF in Scandinavian, directing special attention toward Old Swedish. Here, heads as well as phrases can be stylistically fronted, but in clauses with pronominal subjects, only heads can be targeted. This particular context serves as a fruitful testing ground for conjunct size, more specifically when SF and asymmetric coordinate extraction conspire. Whether such extraction constitutes an ATB violation (extraction of part of a conjunct) or a CSC violation (extraction of a whole conjunct) is the topic of section 5. Only the former analysis has independent empirical support. However, it can be questioned, given what is known about certain prosodic requirements of coordinate ellipsis; it is also weakened by the fact that no other types of unilateral extraction of conjunct parts occur with SF. The clue to distinguishing between the two alternatives is the locus, as it were, of SF. If we assume that this fronting takes place after Spell-Out, an assumption that is independently supported, the whole-conjunct approach is the only one possible: in PF, coordinate asymmetries would be irrelevant, unlike the demand for prosodic licensing. I thus conclude that heads may indeed be conjuncts. Section 6 summarizes the findings of the article.

In this section, I review coordinate structures that have been (or could be) claimed to consist of head conjuncts. First, I discuss Borsley’s (2005) critique of Kayne’s (1994) account of object sharing in coordination (section 3.1); then, I briefly consider coordinate compounds (section 3.2). In neither case is the head analysis called for.

Although the example in (3a) at first appears to involve the coordination of two verbal heads ([_V⁰ criticized], [_V⁰ insulted]) sharing an object ([_DP his boss]), the structure must, according to Kayne (1994:61) “include an empty category”; see (3b).

Kayne’s conclusion is forced by his Linear Correspondence Axiom, which is incompatible with heads being conjuncts (1994:59). Thus, in (3b), there is a phrasal conjunct comprising a verbal head and an invisible object ([_VP criticized [e]]). Kayne’s assumption of an empty category is criticized by Borsley (2005:471), who takes “empty category” to mean deletion. If an example such as (4a) is derived through deletion of the first object (see (4c)), (4a) should mean the same thing as (4b), where both objects are explicit. But, as Borsley notes, “the meanings are different” (2005:472).

Borsley’s critique presupposes a very specific interpretation of the notion of empty category. However, it is not clear why an empty nominal slot in a coordinate structure would always represent a deleted full DP. In fact, evidence from Swedish suggests that there must be different types of empty elements in coordination, depending on the nature of the shared element. As shown in (5)–(6), the sharing of an object between two verb forms can come about in different ways. In contexts where an object in the first conjunct can be coreferential with a pronoun in the second conjunct (as in (5a)), a shared object can follow either the first (see (5b)) or the second conjunct (see (5c)). In contrast, when it is not possible to use such a pronominal object (as in (6a)), a shared object needs to follow the second conjunct (see (6c);¹ cf. the ungrammatical (6b)). I will use the label pro to refer to the empty elements that covary with pronouns (such as the empty object after läste ‘read.PST’ in (5b)).² Note that none of the examples involving a shared or pronominalized object (i.e., (5a–>c), (6c)) has the same meaning as a corresponding example containing an explicit full DP (see (5d), (6d)).

Let us consider the coordinate structures in (5c) and (6c), where the two verbs share a final object. If an empty element is present in the first conjunct (as Kayne (1994) assumes), what is the precise nature of this element? As noted, simple deletion is not tenable. And the element cannot be pro, since sharing of unpronominalizable elements is indeed possible in this context (as in (6c)). In Magnusson (Petzell) 2007:301, I assume that when unpronominalizable elements are shared in coordination, they always correspond to traces in the conjuncts involved. Such an assumption in fact makes the correct prediction that (6b) is bad (see (7a)) and (6c) is good (see (7b)). In (7a), the antecedent få böcker ‘few books’ is too deeply embedded in the initial conjunct to c-command its trace in the second conjunct. In (7b), on the other hand, the final placement of the shared object is derived by remnant movement of the entire coordinate structure from where få böcker has first been symmetrically extracted, providing a configuration in which all traces are properly c-commanded.

Even when the shared element is pronominalizable, as in (5c), the structure in (7b) is valid; see (8a). Pro is certainly semantically compatible with en bok ‘a book’ (i.e., they can be coreferential), but it needs to find its reference in the preceding context, which excludes the analysis in (8b).

To sum up, all of the types of object sharing addressed in this section have been analyzed as elliptic coordination, but with different sorts of null elements that account for the different distribution displayed by different sorts of shared elements. To simply assume that coordination of heads is licit, as Borsley (2005) implies it is, would not help us understand this contrast. The head approach certainly derives both (5c) and (6c), but it offers no explanation for the difference between (5a–b) and (6a–b).³

A remnant movement analysis would be applicable not only to the sharing of an object at the right edge of the sentence (RNR) discussed above, but also to coordinate compounds, exemplified in (9). As Arstein (2005) points out, this type of split compound coordination can have a distributive reading (Paul is a neurochemist and Jane a biochemist) that a pure deletion analysis cannot handle; see example (10), which would be the source of deletion but which has another meaning (each has two degrees).

An analysis of (9) in terms of head coordination is compatible with this semantic discrepancy between (9) and (10), since structural identity is clearly not presumed. However, if (9) represents an elliptic structure created by remnant movement of the entire coordinate complex (as shown in (11)), then (9) and (10) are structurally different as well, which explains the semantic discrepancy.

In other words, head coordination is not the only way to derive a distributive reading in coordinate compounds.

In this section, I first give a brief overview of SF in Old Swedish and Modern Icelandic (section 4.1).⁴ I show that the two varieties differ in one crucial respect: in Old Swedish, but not in Modern Icelandic, there are contexts where the target of fronting cannot be a phrase, only a head. This difference is formalized in section 4.2. I also consider SF of elements that are contained in a coordinate structure, concluding that the fronted element can indeed be a head, but that it is not evident whether the head represents an entire conjunct or just part of one (section 4.3).

In Modern Icelandic,⁵ finite clauses where the subject position is empty (typically embedded clauses) may contain an element just before the finite verb; examples of such fronting, commonly referred to as stylistic fronting (SF), are shown in (12). Either there is no explicit subject at all, as in the relative clauses in (12a–b), or there is a clause-initial subject, such as the interrogative pronoun in (12c). Which element is fronted when there are multiple available candidates has attracted the interest of several scholars (starting with Maling (1980) for Icelandic, followed by Pettersson (1988) and Falk (2007) for Old Swedish). I will not pursue that matter here.⁶ For present purposes, it suffices to conclude that the fronted element can be a phrase (as with the PP in (12b)), but it can also be a single-word expression that could represent a head (i.e., a head-like element), as with the fronted verbs in (12a,c).

But how can we exclude the possibility that what looks like a fronted head (i.e., [_V⁰ buið] in (12a) and [_V⁰ sullað] in (12c)) is in fact a fronted phrase from which everything but the head in question has moved in a previous step of the derivation (i.e., [_VP buið t] and [_VP sullað t])? Such a remnant movement account of SF has been proposed by Franco (2009) and Ott (2009) (apparently independently of each other). Thráinsson (2007:368n15), on the other hand, takes a general stand against any analysis that involves remnant movement, claiming that such analyses make it very difficult to distinguish between types of movement. Shortly, I will argue that the all-phrasal account of SF does not hold. It seems a bit categorical, however, to disregard any remnant movement approach solely on the grounds invoked by Thráinsson (2007); as Bański (2002) shows for Polish, different sorts of movement, although superficially targeting the same string, could be differentiated by different prosody. Also, as Fanselow (2002) argues for German, remnant movement may well occur but is perhaps more restricted than first appears to be the case: Fanselow thus questions the previously well-established analysis of German participle fronting as a case of remnant VP fronting (first suggested by Den Besten and Webelhuth (1990); see (13a)), concluding, however, that such fronting does occur elsewhere (deriving, for instance, (13b)).

Let us now return to the all-phrasal account of SF suggested by Franco (2009) and Ott (2009). Given that unambiguous phrases such as [_PP í Danmörku] can be the target of SF, how can we know that fronted single-word (i.e., head-like) elements are not phrases where nothing but the head is visible? Apparently, head-like and phrasal fronting occur in the same contexts. The only way to exclude an all-phrasal approach would be to find a context where, for some independent reason, it is not tenable to treat the fronted element as an (elliptic) phrase.

Contrary to most descriptions of SF in Modern Icelandic, Hrafnbjargarson (2004b) maintains that it is indeed (at least marginally) possible to have SF even in clauses with a filled subject position. In other words, the requirement that there can be no explicit subject if SF is to take place would not be absolute. However, it is not the case that any explicit subject is licit. It can only be a phonetically weak pronoun. Furthermore, marginal acceptance can be achieved only if the fronted element is a head-like one (as in (14a)); a fronted phrase is completely ungrammatical (see (14b)).

Hrafnbjargarson interprets this discrepancy as an effect of phrasal fronting and head fronting targeting different positions. Most accounts of SF treat the complementary distribution of overt subjects and fronted elements as an indication that subjects and fronted elements occupy the same syntactic slot in some sense (see Thráinsson 2007:368 and the references cited therein; also see the overview in Falk 1993:185–188 and Angantýsson 2011:181ff.). In Hrafnbjargarson’s analysis, this slot corresponds to a specific projection into which subjects and stylistically fronted elements are assumed to move. With no subject present (as in (12)), there can be fronting to the head position (see (12a,c)), as well as to the specifier position of this projection (see (12b)). But if there is a pronoun in the specifier (as in (14a)), there is certainly room for a head, whereas phrasal fronting is predicted to be illicit. In other words, this appears to be the context we are looking for in order to distinguish phrasal fronting from head fronting: if it works with a pronoun, it is a head; if it does not, it is a phrase.⁷ Still, speakers of Icelandic disagree about whether (14a) is marginally acceptable or in fact as ungrammatical as (14b) (see Sigurðsson 2010:177n27). And if the empirical basis is a bit shaky, it seems that the remnant movement account (Franco 2009, Ott 2009) remains an option after all.

However, in Old Swedish, precisely the contrast in (14) appears to have been real. But before we proceed to that, we need to consider some basic SF data. First, SF occurs in clauses with an empty subject position (just as in Icelandic); see (15)–(16). In (15), apparently phrasal elements (such as the direct objects in (15a–b), the predicative in (15c), and the indirect object in (15d)) are stylistically fronted. In (16), by contrast, there is fronting of what looks like single heads (such as the single-word objects in (16a–b), the predicative adjective in (16c), and the nonfinite verb in (16d)); for a more detailed descriptive overview, see Delsing 2001 (cf. also Falk 1993: 179ff.)).⁸

In addition, as already hinted, we find examples that are reminiscent of the Icelandic (14a); consider the examples in (17), where head-like elements are fronted in clauses with pronominal subjects: nonfinite verbs in (17a–b), a pronominal object in (17c), and a predicative participle in (17d).⁹

Still, we lack the corresponding ungrammatical example (cf. (14b)) needed to completely rule out a phrasal approach to (17). Since we cannot, of course, test the intuition of native speakers, we need to look for indirect evidence. Let us therefore turn to the domain of object-verb (OV) order, which I have investigated in a previous article (Petzell 2011). There, I claim that in Old Swedish it was highly dispreferred to move an object over several verbs; see (18a), where the object (rættæ korn tyund ‘right grain tithes’) has moved to the left of both the nonfinite main verb (giöra ‘do.INF’) and the finite auxiliary (wil ‘wants’). Instead, simplex movement, where the object surfaces to the left of the lower verb, was favored; see (18b). Looking more closely at the corpus used in Petzell 2011, there are 71 examples displaying the order in (18b') and only 11 displaying the order in (18a').

The OVfVnf category in Petzell 2011 includes no head-like objects that precede the verbal cluster, either in subjectless clauses (as in (16a–b)) or in clauses with a pronominal subject (see (17c)); I instead analyze these as cases of SF (Petzell 2011:166). In fact, the SF analysis may be applicable in more cases. In Petzell 2011, I do not take into account the fact that all the remaining 11 OVfVnf cases are subjectless. In such contexts, however, SF can target phrases, as we have seen (see (12b), (15a–d)). Consequently, all examples of OVfVnf order in the Old Swedish sample may represent SF of phrasal objects. Taking this generalization into account, there is no need to talk about preference at all; the Old Swedish distribution of preverbal phrasal objects is simply the effect of two different constructions cooccurring: OV, which occurs only within a simplex VP but with all types of subjects (see (18b)), and SF of phrasal objects, which is unaffected by verbal boundaries but instead restricted to clauses without an explicit subject (see (18a)).

Let us now turn to the word order in (19), where a phrasal object occurs before the entire verbal complex, even though the clause has an overt subject (unlike the subjectless (18a)). This example comes from Early Modern Swedish, which had a different OV system from Old Swedish. In Early Modern Swedish, it was preferable to move the object as far to the left as possible, thereby creating examples of this sort (Petzell 2011:174). There are 31 cases of Early Modern OVfVnf in the corpus used in Petzell 2011 (disregarding head-like objects as before); 25 of these have an overt subject, and 10 of them contain, in addition, an unambiguously phrasal object (as in (19)).¹⁰ The dominant order in Old Swedish (i.e., VfOVnf), still occurs in the Early Modern texts but is slightly less common (28 examples) than OVfVnf.

Now, if SF always involves movement of phrases, as Franco (2009) and Ott (2009) suggest, examples like (19), involving phrasal fronting and a pronominal subject, are expected to exist in a variety that allows SF in clauses with a pronominal subject, as Old Swedish apparently does (see (17a–d)). However, examples like (19) are not found in Old Swedish. Certainly, there are cases were more than a head-like element comes between a pronominal subject and the finite verb, but they are (unlike (19)) finite verb–final; see (20a). Verb-final clauses were possible in Old Swedish with any subject, as can be seen in (20b). Consequently, although verb-final clauses may indeed look like SF clauses—namely, when there is only one nonsubject before the finite verb (as with the phrasal object in (20a))—all verb-final contexts need to be disregarded, since there is always the possibility that they are structurally parallel to (20b) (see also Falk 1993:170).

In sum, the conclusion we must draw from the lack of Old Swedish equivalents to (19) is that the examples in (17) do indeed involve head movement; see the analyses of (17a–d) in (21).

This conclusion, in turn, means that the remnant movement approach to SF cannot be universally valid.¹²

Platzack (1988) proposes that the subject pronoun in clauses with SF (as in (17a–d)) cliticizes to the complementizer, in effect creating a clause without a subject, which is a prerequisite for SF. This is shown with example (17a) in (22a). Platzack provides orthographic evidence to support the claim that cliticization to C was indeed possible in Old Swedish; see (22b–c), where there is a reduced form of the masculine pronoun han directly after the complementizer (aen in (22b)) and another reduced form attached to the finite verb (-an in (22c)); also see Alexiadou and Fanselow 2002:239–240, where the authors adopt Platzack’s analysis to be able to explain the loss of V-to-T movement in Mainland Scandinavian.

But the clitic account is unable to explain the lack of phrasal fronting in such contexts on its own. In addition, something like Hrafnbjargarson’s (2004b) analysis is therefore needed, relating the possibility of SF in these cases to the relative lightness of the subject—that is, to the notion that a pronoun is more on a par with a null subject, in some respects, than with a full-fledged DP subject. Let us combine the two approaches into an account of SF that derives this type of construction both in subjectless clauses and in clauses with pronominal subjects (as in Old Swedish), but that also leaves the door open for the latter clause type to be excluded (as in Icelandic).

Suppose that the pronominal subject does cliticize to the complementizer (as Platzack (1988) argues), but that prior to cliticization it has moved to the same phrase as stylistically fronted elements do, located above the canonical subject position (in TP) just below the complementizer (as Hrafnbjargarson (2004b) suggests; also see Bošković 2001). In (23), this phrase is descriptively labeled SP (S for stylistic fronting):¹³ in (23a), the pronoun moves to the specifier position of SP, where it cliticizes to the complementizer, in effect stripping the clause of an overt subject (see (23b)). There is now room for a fronted head (as in (23c)), but not for a phrase (see (23d)), since the only place for a phrase (i.e., the specifier) has already been occupied by the subject clitic (here indicated by the elimination of the maximal SP level).¹⁴

SF in subjectless clauses would have the general structure given in (24a), where SP can harbor either a phrase or a head, but not both (see (24b)). The reason for this restriction is presumably the same as the reason that only cliticizable material may enter Spec,SP if a head is subsequently to be fronted (cf. the nonclitic subject in (24c)), and also the same reason that Spec,TP may not contain overt material (as in (24d)). What, then, unites all the illicit structures in (24)—that is, (24b–d)? Adopting a slightly modified version of Bošković’s (2001) explanation for the empty subject requirement, we could rule out the unwanted output on the grounds that these examples fail to keep the stylistically fronted element adjacent to the finite elements of the clause, that is, both the complementizer and the finite verb.¹⁵ In (24b), the double fronting prevents such adjacency, and in (24c) and (24d), the subject cuts off the link between the fronted element and the complementizer, (24c), and the finite verb, (24d). By contrast, in (23c), where the subject pronoun is attached to the complementizer, and in (24a), where there is no overt subject at all, the adjacency requirement is met.

Why such adjacency should be required is of course an intriguing question, but I will not dig deeper into the matter here. Still, the account is more than just a statement of the facts. First, it felicitously predicts that there will be no SF in nonfinite clauses, as shown in (25).

Second, it predicts that SF should not care, as it were, about overt subjects, as long as they do not block the required adjacency. The Old Swedish example in (26a) appears to lend support to this prediction. What we have here, according to Falk (2007:94–95), is a relative clause with a postverbal subject, where the negation is stylistically fronted, surfacing between the finite verb and the complementizer. In Falk’s analysis, the relative scarcity of examples of this kind is explained by the fact that the subject is itself part of and also highest in the hierarchy of elements that can undergo SF, which, in most cases, leads to the generation of SV order. However, the clause in (26a) represents a violation of the SF hierarchy, Falk argues, targeting the element that is just below the subject on the ladder, namely, negation (2007:96). The order in (26a) would thus be parallel to other hierarchy violations, such as the one in (26b), where the direct object is stylistically fronted over an indirect object. In the majority of cases, it is the other way around; that is, the indirect object is above the direct object in the hierarchy, just as the subject is above negation. Falk’s idea that the subject is the top candidate for SF could certainly be incorporated into my analysis, where SP, as shown above, is a possible landing site for subjects. I will not go into details here, however, since doing so would add nothing significant to the argumentation. Neither will I try to formalize the possibility that the SF hierarchy can sometimes be violated (but see footnote 10).

Let us, instead, return to the proposed adjacency requirement on SF. This requirement is clearly met in (26a), since the subject (albeit overt) comes after the finite verb, thus not interfering between the verb and the fronted element. Such relative inversion (i.e., VS word order in a relative clause) was possible in older Swedish, but is not an option in modern Scandinavian, including Icelandic, which explains why there are no Icelandic counterparts to (26a). As I show in Petzell 2013, allowing such VS order has nothing to do with SF per se, but is linked to the fact that in earlier stages of Swedish (and possibly other older Scandinavian varieties), the canonical subject position was lower than it is today (see also Magnusson (Petzell) 2007 and Håkansson 2008).

In sum, the crucial difference between Icelandic SF and Old Swedish SF is the application of pronominal cliticization (as in (23b)); as we have seen, pronominal SF is, at best, only marginally acceptable to some native speakers of Icelandic (cf. (14a)), indicating that most Icelandic speakers have the structure in (24c) or (24d) for pronominal and nonpronominal subjects alike. Why (23b) is not possible in Icelandic is, of course, an intriguing question. However, pursuing it lies beyond the scope of this article.

Both Icelandic and Old Swedish SF can target elements originating in a coordinate structure, as shown in (27). Old Swedish (as illustrated above) allows SF in subjectless clauses, as well as in clauses with an overt pronominal subject. Coordination-internal elements may be fronted in both these cases: thus, the first of two conjoined verbs is fronted in both (27a), where the subject is a clause-initial relative pronoun, and (27b), where the subject position is filled by the personal pronoun wi ‘we’ (see also the parallel example in (2)). In Icelandic, on the other hand, only subjectless clauses are possible arenas for fronting (disregarding the highly marginal—and, for many speakers, totally ungrammatical—example in (14a)); see (27c).¹⁶

As argued in section 4.1, only examples with pronominal subjects (e.g., (27b)) involve unambiguous head fronting; compare the two possible analyses of (27a) and (27c) in (28a–a′) and (28c–c′) respectively, with the single analysis of (27b) in (28b).

However, although the fronted element in (27b) is a head (see (28b)), it is not evident what it leaves behind in the coordinate structure from which it originates. It could be that it does constitute a head conjunct (see (29a)). At this stage, however, we cannot exclude the possibility that the extracted head is only part of a phrasal conjunct in a coordinate structure involving the same sort of sharing to the right (RNR) that was discussed in section 3.1; see (29b) (cf. (7b), (8a)). Here, the conjuncts share an object (the breff oc beuisning som . . . ‘the letters and evidence that ...’) insofar as they both contain object traces. The placement of the object after the second conjunct follows from the subsequent movement of the entire coordinate structure to the left of the extracted object. As a last step, the verbal head of the first conjunct is stylistically fronted.

The second conjunct in (27b) is clearly larger than a single head: the manner adverb granlica ‘thoroughly’ modifies the second verb ouerlæset ‘over.read.PTC’ but not the first one (also see (2), where the same is true for vinlegha).¹⁸ In (29b), there would thus be VP+VP coordination, and in (29a) V+VP coordination. A priori, neither structure is more likely than the other. Indeed, considering that it is the coordinate structure itself that is under scrutiny here, we need to be careful not to be guided by any presupposed idea that balanced coordination (VP+VP) is somehow more natural than unbalanced (V+VP).¹⁹

On the contrary, we get a first indication that (29a) is in fact more likely than (29b) when we consider more SF examples in a coordinate context. As we have seen, the fronted nonfinite verbs in (27) all share an object with the second nonfinite verb. However, nothing visible is stranded in the first conjunct. Looking more closely at the data, we see that such stranding only occurs in Old Swedish when the extracted element belongs to both conjuncts, as in (30a–b). The same restriction holds for Icelandic; see (31a–b).²⁰

Unlike the fronted verbs in (27), which could represent the whole first conjunct, the fronted elements in (30a–b) and (31a) are clearly only parts of conjuncts. Why stylistically fronted parts of conjuncts need to be tied to all conjuncts involved is an intriguing question, to which I will return in section 5.4. For now, suffice it to say that this apparent symmetry restriction on conjunct parts speaks strongly against the analysis in (29b). Here, the verb is a conjunct part that is extracted from the first conjunct, but the presence of data such as (30a–b) and (31a) and the absence of data such as (31b) indicate that such extraction is in fact illicit (at least in this sort of context). Still, to simply conclude that (29a) is the winner and that head conjuncts are thus a reality is a bit premature. In fact, to assume that an entire conjunct is extracted, as in (29a), is quite problematic, as I will explain further in the next section.

In the previous section, I concluded that it appears more likely that the fronted nonfinite verb in (27b) is a fronted head conjunct (see (29a)) than that it is an extracted part of a phrasal conjunct (see (29b)). Judging from what is known about coordinate asymmetries in general, this conclusion is unexpected: extraction of parts of conjuncts has been shown to be a syntactic possibility, whereas whole-conjunct extraction is not attested (section 5.1). However, fronting of part of a conjunct, in the cases at hand, causes crucial problems for the phonological component (section 5.2). The important task, then, is to determine the locus of SF: is it syntax or is it phonology (section 5.3)? Arguing for the latter, I single out the head conjunct account (i.e., (29a)) as the winner. Finally, I offer some remarks on the role of coordination in PF (section 5.4).

Here, we will be concerned with two basic types of coordinate asymmetry in Old Swedish: violations of the first part of the Coordinate Structure Constraint (CSC, formulated by Ross (1967); see (32a)), and violations of the across-the-board (ATB) restriction (in the sense of Petzell 2010;²¹ see (32b)).

As stated above, (27b) involves SF of a head, which consists of either a whole conjunct or part of a conjunct. The former situation violates the CSC; the latter, the ATB restriction. See (33), which is a repetition of (29), accompanied by a characterization of the nature of each of the asymmetries.

In order to fully evaluate the two analyses in (33), we need to know whether the ATB restriction, on the one hand, and the CSC, on the other, could in fact be violated in other contexts. From the coordinate data presented in section 4.3, it seems that SF does not combine with ATB violations. Still, it is only ATB violations that have been attested elsewhere.

Knowledge of coordination in older Scandinavian is quite fragmentary. The only extensive investigation into the matter is my own study of ATB violations in Early Modern Swedish (Petzell 2010, which is based on my dissertation, Magnusson (Petzell) 2007), where I show that coordinate structures could violate the ATB restriction in a variety of ways until the mid-18th century. The oldest texts that I investigate in Petzell 2010 are from the beginning of the 17th century. The Old Swedish data we have been discussing so far are all older than that. There is, however, no reason to believe that the conclusions reached in Petzell 2010, on the whole, would not be valid for earlier stages of the language as well. Preliminary results presented by Falk (2011) indicate that ATB violations were indeed commonplace during the Old Swedish period.

As for CSC violations, no proper investigation of historical Swedish has been made. However, given what is known from comparative studies of coordinate asymmetries, no search for such examples is likely to succeed: true violations of the CSC seem to be universally scarce (if not nonexistent), whereas ATB violations do occur; see Johannessen 1998:234–235 (see also Zhang 2010:23–26).

In sum, there is independent support for ATB violations, but not for CSC violations. Given these findings, the analysis in (33b), where the fronting violates the ATB restriction, would be preferable to that in (33a), where the CSC is violated. In other words, at this point we have no independent support for the tentative conclusion reached in section 4.3 that conjuncts can be heads.

Although (33b) won the last battle, the war is far from won. Is (33b) at all tenable? If it involves the same type of object sharing that was discussed in section 3.1, namely RNR, we indeed expect it to comply with the restrictions with which RNR usually complies. However, what is peculiar about the RNR in (33b) is that the first verbal head moves further to the left as a case of SF. As I stressed in section 4.3, no such unilateral extraction of other conjunct-internal elements (objects, adjectives, etc.) ever occurs. Also, as will be evident below, the object trace is left without a proper prosodic licenser when the verb is fronted.

In her comprehensive study of various types of coordinate ellipsis, Hartmann (2000:106) formulates the prosodic requirements of RNR as follows: “a RNR construction always contains a narrow focus on the contrasting elements immediately preceding the targets.” A typical RNR example is given in (34a). In the movement-based analysis of RNR, the targets correspond to the two traces created when the shared object moves out of the coordinate structure; see (34b). The contrasting elements are the two verb forms hated and loved.

The contrast need not entail opposition (as in (34)). RNR may involve elements that stand in a purely additive relation to each other, as with chopped and fried in (35). This is also the case in the examples discussed in sections 3.1 and 3.2, where single verbs and first parts of compounds both share a sentence-final argument; see the two elements criticized and insulted in (36a) (cf. (3a), (4a)) and neuro and bio in (36b) (cf. (9)).

What the examples in (34)–(36) have in common is that the targets (i.e., the traces) are all preceded by an element that is prosodically prominent and corresponds to a similar but in some respect contrasting element in a previous or subsequent conjunct. Borrowing the terminology of te Velde (2006:312), we could say that the gaps are licensed by a prosodic feature.

By contrast, in (33b), the first target (t_i) is preceded by an element that lacks phonological features altogether (t_j), since the verb has been extracted and fronted. Object sharing should not be able to come about in this sort of configuration, because a proper licenser of the object gap is missing. Therefore, (33b) is not an acceptable analysis of (27b). (33a), on the other hand, contains no object gaps that need licensing, since there is only one object that is the complement of the conjoined verbs.

How should (27b) be analyzed, then? As we have seen, both alternatives (i.e., (33a–b)) appear to be impossible: (33a) violates the CSC and (33b) contains a prosodically unlicensed object gap. How can this be? I propose that the solution lies in the apparent extrasyntactic nature of SF. Holmberg (2000:469) stresses that this sort of fronting in Icelandic is expletive in nature. Indeed, it is applicable in more or less the same clause types as the expletive pronoun það ‘it’ (Holmberg 2000:467, Thráinsson 2007:375ff.). In addition, it has hardly any semantic effect on the clause (Holmberg 2000:466, Thráinsson 2007:370). Both of these characteristics suggest that the movement in question takes place after Spell-Out. In other words, it appears to be a PF operation. This approach has been implemented by Sigurðsson, who maintains that SF is an effect of what he calls the Fill the Left Edge requirement, (2010:184–185), which is a “PF or performance target” (2010:179), different from syntactic movement.

However, as Holmberg notes, the lack of stylistically fronted auxiliaries suggests that there is a semantic side to the matter as well (2000:468). The fact that SF can distinguish between auxiliary and main verbs certainly appears to be problematic for the PF account, as stressed by Egerland (2013). Egerland maintains (drawing on general suggestions made by Erteschik-Shir (2006, 2007)) that it is the assignment of information-structural features (which target only lexical, not functional words) that explains why the auxiliary (a functional word) is not available for SF (2013:72). However, as far as I can understand, Egerland’s account is fully compatible with SF being a PF phenomenon. If SF is indeed triggered by a background feature (or rather, an unspecified information-structural feature), as Egerland suggests (2013:69), this feature would be visible in PF. If such features were not, there would, for instance, be no proper assignment of stress to elements bearing a focus feature.

A PF account of SF clearly offers a promising way out of our deadlock, since it differentiates between (33a) and (33b). The prosodic requirements of RNR must be relevant in PF. Therefore (33b), where these requirements are not met—that is, there is no contrasting focus before the target—must be rejected. On the other hand, the extraction of the stylistically fronted head in (33a) causes no problems: the coordinate structure is kept syntactically intact and is not split until it reaches PF, since the split is motivated by a PF process (i.e., SF). As a consequence, we have no reason to exclude (33a) on the grounds that it constitutes a CSC violation. The beating about the bush regarding the analysis of (27b) has thus come to an end: it must be analyzed as in (33a), where the first conjunct of the coordinate structure is a head.²²

Now, let us address the empty subject requirement of SF in light of the proposed PF analysis. In section 4.2, I analyzed this requirement as a demand for adjacency between the fronted element and the complementizer/finite verb, largely following Bošković (2001). To account for the fact that pronouns do not block the licensing of SF in Old Swedish, I assumed that cliticization precedes the evaluation of adjacency. If SF is a PF phenomenon, PF would be the locus of all the different steps involved: subject cliticization, the actual fronting, and adjacency evaluation. In other words, PF needs to involve movement, and it also needs to be quite complex in the sense that a certain derivational order can apply. Such a conclusion is fully in line with Sigurðsson’s (2010) proposal that PF is larger than traditionally assumed. It is also compatible with recent approaches to PF within Distributed Morphology; see, especially, Kandybowicz’s (2008:11–12) proposal of a layered PF, where processes such as morphological fusion (e.g., cliticization) precede phonological mapping operations (under which the adjacency evaluation would fall). Furthermore, SF is not unique in taking place after Spell-Out: according to Agbayani and Golston (2010), PF movement occurred in Classical Greek, creating split constituents that narrow syntax cannot derive.

Still, it is not evident how the adjacency requirement is to be formulated more precisely. In Bošković’s version, the head of the phrase to which stylistically fronted elements move is an abstract verbal affix, “which must merge under PF adjacency with a verb (more precisely a finite verb given that stylistic fronting cannot occur in infinitives)” (2001:250). If a subject comes between this affix and the finite verb, there can be no adjacency. In my modified account, I propose a requirement for double adjacency in order to account for blocking effects both between the fronted element and the complementizer (see (24c)) and between the fronted element and the finite verb (see (24d)). As a consequence, the affix account cannot be adopted; instead, the demand for adjacency must be directed, as it were, toward the fronted element itself.

Finally, the adjacency account, in both Bošković’s version and mine, presupposes that finiteness can be identified within PF. Exactly how this comes about is not entirely clear. Are finite features indeed visible in this more syntactic Sigurðssonian PF? Or is there something else about the elements at hand that enables PF to single them out? In order to resolve the issue, a more thorough discussion of the nature of PF is clearly needed. I have started such a discussion here, but it will have to be continued elsewhere.

What implications do the findings presented here have for a theory of coordination? In a way, the progression of this article bears a resemblance to losing your keys and then discovering that you had them all along. Although the conjuncts addressed in section 3 could be treated as phrases, the SF story shows that head conjuncts are a syntactic reality after all, in effect ruling out the all-phrasal approach to conjuncts (in, e.g., Kayne 1994 and Johannessen 1998). However, there are necessarily restrictions regarding what heads are eligible for coordination. Otherwise, we would expect all sorts of coordinated clitics to occur, for instance, which is contrary to fact (as the French example (37) shows).

Minimalist approaches to coordination that acknowledge the existence of head conjuncts naturally have to account for this. Zhang (2010) incorporates Chomsky’s (1994) bare phrase structure approach into her theory of coordination, making constituent level (head or phrase) irrelevant; instead, all conjuncts are required to be nonprojecting (Zhang 2010:37ff.). Another proposal is presented by te Velde (2006); in his view, it is conjunctions that are nonprojecting (and in that way similar to adverbs, in the spirit of Travis 1988), simply conjoining individually derived structures (heads as well as phrases) into coordinate complexes.²³

We need not go into the details of these two analyses. Both could easily, it seems, incorporate the crucial Old Swedish coordination involving a raised head conjunct and a second, slightly larger conjunct (as in (2) and (27b)). Neither te Velde’s nor Zhang’s analysis entails the demand for constituent-level symmetry. In other words, the V+VP structures in (33a) can be derived straightforwardly.²⁴ However, both analyses appear to lack some restrictive component that can prevent the system from deriving illicit head coordination (as in (37)), without excluding the head conjuncts that we do want the system to derive.

Te Velde actually addresses the very example shown in (37). Although derived by the syntactic component, it can still felicitously be blocked because clitics lack the prosodic prominence that, in te Velde’s view, all coordinate conjuncts are required to have (2006:116n28). At first, this strategy appears to exclude our SF cases as well: in the syntactic domain, they are complete (see (38a)), but in PF, the coordination is split (see (38b)), leaving the first conjunct as prosodically problematic as the clitic conjuncts in (37); compare (38c).

However, it is not evident what status coordination has in PF. One might, for instance, inquire whether the notion of conjunct is really relevant (as te Velde implies it is). Above, I have argued that coordinate structures must not be split in syntax (CSC violations), but that PF does not care, as it were, about such asymmetries. The reason that PF does not care may be that the coordinate structure is irrelevant in PF, or at least that it is not active in the way it is in syntax. In that case, we can exclude the ungrammatical clitics in (37) without taking the coordinate structure per se into account. A pronominal clitic of the sort occurring in (37) needs a verbal host, and there simply are not enough hosts in (37). (38c) is, by contrast, unproblematic: the first conjunct would contain no phonological features (after SF has applied) and would therefore, naturally, be irrelevant in PF.

Furthermore, if the coordinate structure is deactivated in PF, this may explain the lack of SF of conjunct-internal material. As observed in section 4.3, only elements that are shared by all conjuncts may be stylistically fronted (see (39a), which is a modified version of (30a)). This symmetry restriction is clearly unrelated to the SF. Similar sharing occurs in clauses without any fronting, as can be seen in (39b).

In other words, al höghmäle ‘all high.cases’ in (39a) is not extracted from the coordinate structure, but is external to it (just like the objects in (39b)) when SF applies. Consequently, the only part of a coordinate structure that we need to assume is accessible for fronting is the entire first conjunct. As we have seen, such conjunct fronting occurs both in clauses with pronominal subjects (as in (2), (27b)) and in subjectless clauses (as in (27a,c)). The fronting of a shared complement (as in (39a)), however, occurs only in the latter context. Obviously, a complement is always a phrase, never a head (as shown in (40a), from present-day Swedish). And, as we know, only heads may be fronted in Old Swedish clauses involving pronominal subjects; thus, the lack of complement sharing in such contexts is expected. I illustrate this limitation in (40b) using the same present-day Swedish forms as in (40a) (owing to the obvious lack of negative data).

On the other hand, fronting of a head within a shared complement is predicted to be possible in clauses with a pronominal subject; given that the complement is outside the coordinate structure, its content should (just like parts of complements in the simple clauses in (21)) be accessible for SF. As shown in (41), this prediction is borne out. This example features a pronominal subject (iak ‘I’) and two coordinated passive verbs that share a PP complement: the complement of P is relativized, but P itself is stylistically fronted.

To sum up, SF may target an initial conjunct but not part of it, indicating that the more embedded elements of the coordinate structure are inaccessible for movement at this stage of the derivation. However, it is clearly too drastic to conclude that PF has no access whatsoever to parts of conjuncts. As I argue in Petzell 2010, there appears to be purely phonological matching between conjuncts in PF, which suggests that the internal structure of conjuncts is indeed visible in some sense. Also, a prerequisite for prosodic licensing of coordinate gaps in RNR structures of the sort discussed in sections 3.1 and 5.2 is that the conjunct where the gap resides can be reached and reviewed in PF.

The aim of this article has been to determine whether all coordinate conjuncts are phrases or whether heads can be conjuncts. I have argued that for alleged head conjuncts posited in the literature, a phrasal account is equally as good as (in the case of coordinate compounds) or even preferable to (in the case of object sharing) a head conjunct analysis.

To get to the bottom of the question at hand, I have considered Scandinavian stylistic fronting. Such fronting may target phrases, which makes it difficult to fully rule out the possibility that even when a head-like element occurs, it is in fact a phrase containing nothing but a head. However, in Old Swedish (unlike Modern Icelandic, the variety usually invoked in the literature on SF), there is a context where SF cannot target phrases, namely, clauses with a pronominal subject; here, only head-like elements can be fronted. Since phrases never occur, we are forced to conclude that the head-like elements are indeed heads and not remnant phrases.

Having established that the clauses with pronominal subjects are unambiguous head contexts, I have considered a particular subtype of such cases where the first of two coordinated verbs is fronted. While it is certainly clear that head fronting is going on, it is not clear what the extraction of the verbal head in fact leaves behind: is the fronted verb really a full nonphrasal conjunct, or is it merely part of an elliptic phrasal conjunct? Both possibilities are problematic. I have shown that the latter alternative (phrasal conjunct) is incompatible with the prosodic requirements of coordinate ellipsis; also, it predicts that SF should occur with conjunct-internal elements in general (not only with verbs), which is not the case. The former alternative (head conjunct) instead involves a severe type of coordinate asymmetry, known to be banned from syntax in general. Still, this asymmetry turns out to be irrelevant, since the extraction creating it appears to be extrasyntactic, occurring in PF. Here, however, the violation of prosodic requirements is, naturally, crucial. In sum, the head conjunct analysis is favored.

• Vidh

  [“Västgötalagens lagmanna-, kunga-, och biskopslängd”]. Collin, Hans Samuel, and Carl Johan Schlyter, eds. 1827. Samling af Sweriges gamla lagar. Vol. 1, Westgötalagen. Stockholm: Z. Haeggström. (written about 1240)

• Fleg

  Stephens, George, ed. 1847. Svenska medeltidens kloster-och helgonabok: En samling af de äldste på svenska skrifne legender och äfventyr. 2, Ett forn-svenskt legendarium innehållande medeltids kloster-sagor om helgon, påfvar och kejsare ifrån det I:sta till det XIII:de århundradet (Bd 1). Stockholm: Norstedts. (written 1276–1308)

• ÖgL

  Collin, Hans Samuel, and Carl Johan Schlyter, eds. 1830. Corpus iuris sueo-gotorum antiqui: Samling af Sweriges gamla lagar, [...]. Vol. 2, Codex iuris Ostrogotici = Östgöta-lagen. Stockholm: Norstedts. (written in the 1290s)

• UL

  Henning, Samuel, ed. 1967. Upplandslagen: Enligt Cod. Holm. B 199 och 1607 års utgåva. (Samlingar utgivna av Svenska fornskriftsällskapet 70:1.) Uppsala: Almqvist & Wiksell. (written in 1297)

• Kstyr

  Moberg, Lennart, ed. 1964. En nyttigh bok om konnunga styrilse och höfdinga [facsimile of edition from 1634 by Johannes Bureus]. Uppsala: Almqvist & Wiksell. (written about 1330)

• MEL

  Collin, Hans Samuel, and Carl Johan Schlyter, eds. 1862. Corpus iuris sueo-gotorum antiqui: Samling af Sweriges gamla lagar, [...] Vol. 10, Codex iuris communis Sueciæ Magnæanus = Konung Magnus Erikssons landslag. Lund: Berlingska. (written about 1350)

• Birg

  Klemming, Gustaf, ed. 1857–1858. Heliga Birgittas Uppenbarelser. (Samlingar utgivna av Svenska fornskriftsällskapet 14:1, bok 1–3.) Stockholm: Norstedts. (written after 1380)

• K-M

  Klemming, Gustaf, ed. 1887–1889. Prosadikter från Medeltiden: Aff Karlamagnus konung. (Samlingar utgivna av Svenska fornskriftsällskapet 28.) Stockholm: Norstedts. (written at the end of the 14th century)

• SDHK

  Charters from 1406 (16663, 16673) and 1414 (18212) in Svenskt Diplomatariums huvudkartotek (http://sok.riksarkivet.se/SDHK).

• Birg2

  Klemming, Gustaf, ed. 1860. Heliga Birgittas Uppenbarelser. (Samlingar utgivna av Svenska fornskriftsällskapet 14:2, bok 4–5.) Stockholm: Norstedts. (written at the beginning of the 15th century)

• B-J

  Klemming, Gustaf, ed. 1887–1889. Prosadikter från Medeltiden: Barlaam och Josafat. (Samlingar utgivna av Svenska fornskriftsällskapet 28.) Stockholm: Norstedts. (written about 1440)

• Pkrön

  Klemming, Gustaf, ed. 1881. Sveriges krönika (vanligen kallad den prosaiska). (Småstycken på forn-svenska 1.) Stockholm: Svenska fornskriftsällskapet. (written in the 1450s)

• Gyll

  Hausen, Reinhold, ed. 1882. Diarium Gyllenianum eller P. M. Gyllenii dagbok 1622–1667. Helsingfors. (author born in 1622)