Remnant movement
The fact that complements do not double in the predicate fronting construction, shown in (14), is perhaps the least surprising facet of the predicate fronting puzzle. It is therefore a suitable starting point.
-
(14)
The unacceptability of sentences with two overt complements follows straightforwardly from a theory of syntax that allows for the fronting of incomplete (or “remnant”) categories. Remnant movement is the phenomenon whereby a phrase undergoes movement only after the extraction of some smaller element from it earlier in the derivation. Consider the Yiddish sentence (15), modeled after a German sentence that has been cited as evidence for remnant movement (Müller 1998:ix, ex. 1).
-
(15)
Within the fronted VP, only the V geleyent is pronounced. The intuition is that an earlier operation extracts the DP dos bukh from the VP and generates a silent trace, which remains silent within the fronted VP. The fronted VP is thus a “remnant category” with respect to the DP. The general pattern is schematized in (16), with copies rather than traces.
-
(16)
According to the Inclusiveness Condition (Chomsky 1995:225), a principle of minimalist syntax stipulating that syntactic operations like Merge cannot generate new objects such as silent traces, all occurrences of an element like YP in (16) are syntactically identical and eligible for phonetic realization. Under typical conditions, one might expect Spell-Out to result in the pronunciation of the highest copy of a moved object rather than a middle one,Footnote 10 although it is the middle occurrence of YP that is pronounced here.
In order to capture the intuition behind remnant movement within a minimalist framework that invokes copies rather than traces, I adopt the Spell-Out conditions of Collins and Stabler (2016). Their formalization of minimalist syntax is particularly appealing because it provides the analyst a set of precise definitions with which to evaluate the predictions of novel proposals—specifically, definitions regarding the syntactic occurrences that come to be pronounced in any given derivation. Because their formalization was not designed for Yiddish, and because the authors themselves “put aside issues such as how to handle the copies formed in predicate clefts” (Collins and Stabler 2016:71), the fact that their system can be applied to the Yiddish predicate fronting construction (albeit with one modification) lends support to it.
The Spell-Out conditions are provided in (17) and explained in the paragraphs that follow.
-
(17)
Spell-Out conditions for phrasal movement, including remnant movement (based on Collins and Stabler 2016, definitions 40 and 41-c):
-
a.
X ∈ {X,Y} is final in a syntactic object SO iff there is no Z contained in (or equal to) SO such that Z immediately contains X, and Z contains the set {X,Y}. Otherwise, X is nonfinal in SO.
-
b.
If SO = {X,Y} and X in SO is final in Phase but Y is not, TransferPF(Phase, SO) = TransferPF(Phase, X).
Condition (17-a) specifies the occurrences of syntactic objects (i.e., lexical item tokens or sets of such syntactic objects) that are evaluated to be “final” or “nonfinal” based on their structural positions. Condition (17-b) then states that only “final” occurrences are phonetically realized after TransferPF (Spell-Out) of a phase, the unit to which Transfer applies. The first condition makes reference to the notions “immediate containment” and “containment,” which Collins and Stabler (2016:46) define as follows: “immediate containment” is set membership (i.e., Z immediately contains X iff X ∈ Z), and “containment” includes both immediate containment and transitive containment extending down the tree (e.g., Z contains X if Z immediately contains X, or if Z immediately contains Q which immediately contains X, etc.). In other words, a syntactic object immediately contains each of its daughters, whereas a syntactic object contains each of its daughters, granddaughters, great-granddaughters, etc.
With these definitions in mind, the predictions of the Spell-Out conditions in (17) are illustrated in the following derivations. All occurrences with matching letter labels are assumed to be syntactically identical; the subscripts are for expository purposes only.
-
(18)
In this derivation, A1 is an element of the set {A,B} = C and raises to become the sister to C. According to (17-a), the occurrence A1 in {A,B} is nonfinal within D, because there exists a syntactic object (namely D itself) that immediately contains A (the occurrence A2) and also contains the set {A,B}. The occurrence A2 in {A,C} is final in D, however, because there exists no syntactic object that immediately contains an occurrence of A and also contains the set {A,C}. (D immediately contains an occurrence of A, but D does not contain the set {A,C}; rather, D is the set {A,C}.) B, C, and D are final in D by the same reasoning. Consequently, according to (17-b), transferring D to PF will yield the phonetic realization of A2 and B only, as we would normally expect in simple cases of syntactic movement.Footnote 11
Crucially for our purposes, the two-part definition in (17) also correctly predicts the occurrences that are pronounced under remnant movement. Consider the following derivation (19), which contains the subtree shown in (18). We will assume that E is a phase.
-
(19)
A1 is nonfinal in E because there is an object (D) that immediately contains A2 and also contains the set {A,B}. A2 is still considered final because there is no syntactic object anywhere that immediately contains an occurrence of A and also contains the set {A,C}. Now consider the occurrence A3 in {A3,B2} = C2, the moved remnant category. The occurrence A3 is nonfinal in E because there exists a syntactic object within E, namely D, that immediately contains an occurrence of A and also contains the set {A,B}. It follows from the same reasoning that C2, D, and E are final within E, while C1 is nonfinal. As a result, transferring E to PF will yield overt occurrences of B2 and A2 only.
In this way, the Spell-Out conditions of Collins and Stabler (2016) correctly predict the pronunciation of occurrences under typical phrasal movement as well as remnant phrasal movement, without relying on the generation of silent traces.
Remnant movement across a phase
One complication of this approach to pronunciation under remnant movement relates to derivation by phase. A fragment of the derivation of the Yiddish remnant movement sentence in (15) is shown in (20). Again, all occurrences of a given object are syntactically identical and numbered for expository purposes only.
-
(20)
In order for the object DP dos bukh ‘the book’ to be pronounced above the subject DP yeder yidishist ‘every Yiddishist,’ and in order for the VP geleyent <dos bukh> ‘read <the book>’ to eventually move to the left periphery, both of these phrases must move through the edge of the vP phase. This is because TransferPF applies to phases (17-b), and if one adopts the Phase Impenetrability Condition (PIC; Chomsky 2001:13), then any objects which remain in the domain (complement) of v become inaccessible once they are spelled out. This inaccessibility could be due to the removal of syntactic structure, or to a stipulation that such structure is invisible to higher operations (Chomsky et al. 2019:240–241). In either case, this inaccessibility creates a “lethal problem” for the pronunciation of occurrences (Collins and Stabler 2016:73): if Spell-Out renders the base-generated VP in (20) inaccessible to the condition that determines the finality of occurrences (17-a), then the two higher occurrences of the DP dos bukh (marked #2 and #3) are both predicted to be final. DP dos bukh2 is final in vP because there is no object that immediately contains an occurrence of the DP and also contains the set {DP dos bukh,[middle] v′}. DP dos bukh3 in the fronted VP is also final, because the only syntactic object that immediately contains an occurrence of DP dos bukh (namely the highest v′) does not contain the accessible set {V geleyent,DP dos bukh}. The “lethal problem” for remnant movement is therefore the doubling of the object, whenever the object and the VP move to positions out of the phase in which the VP was base-generated.
This problem—that “if an occurrence A were nonfinal before Transfer, it may become final after Transfer” (Collins and Stabler 2016:73)—leads the authors to sketch out (in an unpublished Appendix) an alternative conceptualization of Transfer that preserves syntactic structure while rendering it inaccessible to operations like Merge. However, in this article I adopt the strict assumption that any lower structure that has been spelled out in a phase is completely invisible for the determination of final and nonfinal occurrences at a higher phase. If this strict version of the PIC is to be maintained, then the derivation of remnant movement that causes doubling (20) must be replaced by a more optimal (less costly) alternative that does not cause doubling. A minimally different derivation is one in which the landing sites for the DP and remnant VP—the inner and outer vP specifiers, respectively—are reversed, as in (21):
-
(21)
Here, the DP dos bukh3 is final in vP because there is no object that immediately contains an occurrence of the DP and also contains the set {DP dos bukh,[highest] v′}. DP dos bukh2 is nonfinal in vP because there is an object (namely vP itself) that immediately contains an occurrence of the DP and also contains the set {V geleyent,DP dos bukh2}. (Crucially, neither of these calculations requires the VP complement of vP to be accessible.) When the remnant VP fronts to the left periphery in Spec,TopP (not shown here), the DP dos bukh contained within that left peripheral VP will also be nonfinal, because both Spec,vP positions will still be accessible at the TopP phase. This derivation thus yields correct predictions about remnant movement, if one adopts the Spell-Out conditions (17) and the strict assumption that spelled out structure is completely invisible for the purposes of calculating the finality of occurrences.
The structure shown in (21) could be derived in a few different ways. One would involve movement of the DP dos bukh to Spec,vP followed by movement of the remnant VP to an inner specifier, as in the “tucking in” approach of Richards (1999). Tucking in the VP below the landing site of the DP is difficult to define within the framework of Collins and Stabler (2016), since it requires that the vP object formed by movement of the DP be dismantled and subsequently reconstituted with an intervening specifier (see Collins and Stabler 2016:59, fn. 9). Another way to derive (21) is to assume that the VP moves before the DP does. For example, the VP might raise to an inner Spec,vP and the DP contained within it might raise to an outer Spec,vP. This is reminiscent of smuggling, where movement of a phrase XP containing YP precedes the evacuation of YP from XP (Belletti and Collins 2021:3).Footnote 12 However, it is not clear what could motivate the extraction of a syntactic object from an inner to an outer vP specifier (e.g., according to theories of Agree).Footnote 13 Alternatively, the v head could bear features that trigger movement of both the DP and the VP; arguably it is the VP which is attracted first, because it is closer than the DP, giving rise to the structure shown in (21) and ruling out the structure shown in (20). While I remain agnostic about which of these approaches is correct, the derivations in this article will include movement arrows suggestive of the smuggling approach. In any event, I proceed with the assumption that the DP and remnant VP escape the vP phase by landing at specifiers in this specific configuration, such that the DP c-commands the VP.
This discussion is not meant to redefine “remnant movement” so that it always involves three instances of movement (smuggling of XP, extraction of YP from XP, and movement of XP) rather than two (extraction of YP from XP and movement of XP). In (21), we would still say that “remnant movement” involves extraction of the DP dos bukh from the VP and subsequent fronting of the remnant VP to the left periphery (not shown). Smuggling of the VP is thus a separate step that enables correct pronunciation when remnant movement involves occurrences across multiple phases.
Remnant VP movement and predicate fronting
If the Spell-Out conditions in (17) are part of Universal Grammar, it follows that predicate fronting in Yiddish should never give rise to doubling of complements (the data are repeated in (22)).
-
(22)
If the complement DP mame-loshn ‘mama-language’ is able to undergo object shift to some intermediary position prior in the derivation to the fronting of the VP, then the occurrence of the DP contained in the fronted remnant VP will not be pronounced (yielding (22-a)). On the other hand, only if object shift does not occur is it possible, given these conditions, to pronounce an occurrence of the DP in the fronted VP (yielding (22-b)). This is essentially the analysis provided by Abels (2001) for predicate fronting in Russian and by Bondaruk (2009) for predicate fronting in Polish, in which a fronted “bare verb” is the result of remnant verb phrase movement after the evacuation of its complement.
In order to account for the variation in where the complement is pronounced—it can be realized after the first or second copy of the verb—it is assumed that object shift itself is optional, consistent with the proposals of Johnson (1991:606) and Lasnik (2001).Footnote 14 Independent evidence for the optionality of object shift in Yiddish comes from the following alternation:
-
(23)
Sentence (23-a) suggests that the object DP dem artikl is free to remain in its base-generated position to the right of the verb (assuming Yiddish is SVO; Diesing 1997). The acceptability of (23-b) suggests that there must be a relatively low projection that can host shifted objects.Footnote 15 Finally, sentence (23-c) shows that the object can scramble past negation (presumably through Spec,vP).Footnote 16
While the conditions under which the movement of the object occurs or does not occur are beyond the scope of the current study (see Diesing 1997 for discussion), it is important to note that weak pronominal objects differ from other definite objects in that they must move to a position above negation (Diesing 1997:393–394; see Thráinsson 2001 and Vikner 2006 for parallels in other Germanic languages). This observation will become relevant shortly.
-
(24)
We can see how the Spell-Out conditions in (17) would apply in the predicate fronting construction to yield the pronunciation of just one copy of the object—after the first verb or after the second, but not in both places. Consider the sentence in (25-a), where curly braces indicate that the object must be pronounced in just one of two places. A fragment of the derivation of this sentence is shown in (25-b). Optionality of object shift is indicated with a dashed arrow, a convention adopted throughout the rest of the article.
-
(25)
If the VP pronounced in the left periphery is the result of movement (as argued in Sect. 2), then it must move through the edge of the vP phase in order to be accessible for further movement. If object shift to a high projection above negation is optional, presumably the DP must also pass through the edge of the phase. (The object’s ultimate landing site beyond the verb’s extended projection is not shown here, and its label is not important for our purposes.)
In the derivation shown in (25-b), the VP raises to Spec,vP, rendering the lower occurrence nonfinal (and silent) at Spell-Out, at which point it becomes inaccessible to higher operations. Assume that object shift (the dotted arrow) has occurred. The Spell-Out conditions in (17) predict that the occurrence DP dos bukh3 is final in vP. This is because there is no syntactic object anywhere that immediately contains an occurrence of the DP dos bukh and also contains the set {DP dos bukh,[highest] v′}. (Of course, if the object moves to some higher projection, e.g., above negation, the Spell-Out conditions predict that the object will be final and pronounced there rather than in Spec,vP.) The occurrence DP dos bukh2 is nonfinal; this is because within vP, there is a syntactic object (vP itself) that immediately contains the occurrence DP dos bukh3 and also contains the (accessible) set {V leyen,DP dos bukh2} in its inner specifier.
However, if object shift does not occur—so that there is no occurrence DP dos bukh3—then every occurrence of the DP is contained within an occurrence of VP. Without object shift, the occurrence DP dos bukh2 is final in vP. If the higher VP subsequently fronts to the left periphery, then the left peripheral VP will be final in TopP and it will contain the only pronounced occurrence of the DP object.
One of the predictions of this remnant movement approach to complements is that whenever object shift is obligatory, it should not be possible for the complement to be pronounced in the fronted VP. For example, because the movement of pronominal objects is obligatory (24), subsequent fronting of the VP should always be a case of remnant movement; thus the fronted VP should always contain a nonfinal (silent) occurrence of the pronominal object. This prediction is confirmed by my consultants:Footnote 17
-
(26)
As this discussion has shown, the formal Spell-Out conditions of Collins and Stabler (2016) yield the predictions of remnant movement, and therefore the obligatory single copy pronunciation of DP complements in the Yiddish predicate fronting construction. In order to evaluate whether the Spell-Out conditions make correct predictions with regard to occurrences of the verb—that the verb is obligatorily pronounced in two places—we first need to ensure that the minimalist framework of Collins and Stabler (2016) can handle simple instances of head movement.
Spell-Out conditions and head movement
The formal conditions specifying which occurrences of syntactic objects come to be pronounced at PF (17) refer neither to phrases nor to heads. As demonstrated in the previous section, they correctly predict the pronunciation of phrasal constituents. However, it turns out that the conditions are incompatible with head movement. In fact, Collins and Stabler (2016:43) specifically mention head movement as one of the topics not addressed in their formalization due to space limitations. We will therefore need to modify Collins and Stabler’s (2016) Spell-Out conditions to make correct predictions for the pronunciation of occurrences of syntactic heads before they can be applied to the verbal occurrences in the predicate fronting construction.
To demonstrate why these conditions on Spell-Out are incompatible with head movement, consider the simple case of V-to-v movement shown below in (27) for the Yiddish verb phrase red mame-loshn ‘speak Yiddish’ (specifiers not shown).
-
(27)
If we adopt the assumption that head movement is syntactic and involves the formation of complex head adjunction structures,Footnote 18 then applying the definition of “final” (17-a) to the occurrences of V in (27) gives rise to an unexpected result. The left occurrence of V red is final, because there exists no syntactic object that immediately contains an occurrence of V red and also contains the set {V red,v}. However, the right occurrence of V red is also final, because there exists no syntactic object that immediately contains an occurrence of V red and also contains the set {V red,DP mame-loshn}. As a result, the definition given in (17-b) predicts that two copies of the head red will be pronounced. In fact, the Spell-Out conditions predict that every instance of successive-cyclic head movement will give rise to a final, and thus phonetically overt, occurrence of the moved head. This is clearly an undesirable result, since head movement does not normally give rise to the pronunciation of multiple copies.
To avoid this problem, we appeal to the observation that adjunction involves the formation of syntactic objects consisting of multiple segments: [β α [β …]] (Chomsky 1986:7). Chomsky has posited an operation, Pair-Merge, to formalize such a multi-segment category whenever one object is adjoined to another:
Adjunction has an inherent asymmetry: X is adjoined to Y. Exploiting that property, let us take the distinction between substitution and adjunction to be the (minimal) distinction between the set {α,β} and the ordered pair <α,β>, α adjoined to β… For clarity, let us refer to substitution as Set-Merge and adjunction as Pair-Merge. (Chomsky 2000:133)
The notion that Pair-Merge forms adjunction structures has been repeated in Chomsky’s subsequent writings (e.g., Chomsky 2004:117–118; Chomsky 2008:146–147) and has also been assumed to apply to head movement (Chomsky 2015:12). We define Pair-Merge as follows:
-
(28)
Pair-Merge for adjunction:
Pair-Merge(X,Y) = <X,Y>
The most straightforward way to accommodate head movement within the Spell-Out conditions in (17) is to expand the definition of “immediate containment” to apply even in the presence of an intervening segment of a two-segment category, formed when X is adjoined to some other syntactic object Q:
-
(29)
Immediate containment:
Z immediately contains X iff X is a member of Z or <X,Q> is a member of Z.
Note that Collins and Stabler (2016:43) specifically name “Pair-Merge (adjunction)” as a topic that, like head movement, is not addressed in their formalization due to space limitations. The definitions in (28) and (29) are thus one way to fill a gap in their framework.
Returning to the head movement derivation in (27), we can already see the effect of this modification. The externally merged V red is nonfinal in vP because there exists a syntactic object (namely vP itself) that immediately contains <V red,v> and also contains the set {V red,DP mame-loshn}. The complex v head is final because there is no object that immediately contains v and also contains the set {v,VP}. In this way, the next higher head “counts” as an occurrence of the immediately lower head and ensures that the lower head is nonfinal.
Once again, a fragment of the derivation of the sentence (25-a) leyen-en {dos bukh} leyen ikh {dos bukh} ‘As for reading {the book}, I am reading {the book}’ is shown in (30)—this time with verb movement to T and the addition of complex head adjunction structures. Here we focus only on the occurrences of V leyen ‘read’ that are involved in head movement up the tree (i.e., occurrences #1, #2, and #4).
-
(30)
The externally merged verb in {V leyen1,DP dos bukh} is nonfinal in the vP phase: there is a syntactic object (the lowest v′) that immediately contains <V leyen2,v> and also contains the set {V leyen1,DP dos bukh}. Because V leyen1 is nonfinal, when TransferPF applies to the phase it will be silent. The strict assumption of the PIC also means that once the lower VP is transferred to PF it is no longer visible for the purposes of calculating the finality of occurrences higher in the tree.
The next higher head, the complex v1, is also predicted to be nonfinal in TP: this is because T′ immediately contains <v2,T> and it also contains the set
(strikethrough represents that the VP is inaccessible after phase transfer). The highest complex head, T, is final in TP: there is no syntactic object that immediately contains T (or any <T,Q>) and also contains the set {T,vP}.
As the verb moves successive-cyclically up the tree, forming increasingly more complex head adjunction structures, each complex head thus renders the next lower head nonfinal. T is the highest complex head, and it is the only one that is final. While this update to “immediate containment” makes correct predictions in simple cases of syntactic head movement, the analysis obviously does not address the classic problem of head movement’s violating the Extension Condition (Chomsky 1995:190–191), since head adjunction does not extend the tree at the root. However, as a rule on Spell-Out, it is both consistent with the spirit of the syntactic framework of Collins and Stabler (2016) (where finality is determined solely by structural configurations) and descriptively adequate, as it allows for a complex head to “count” as an occurrence of the next lower head and thereby render it nonfinal.
With this update in mind, we can return to the second puzzle of the Yiddish predicate fronting construction: the obligatory doubling of verbs. As will be demonstrated, the doubling of verbs actually does not require any additional modifications to the Spell-Out conditions.