1. INTRODUCTION
This article examines A′-movement, such as wh-movement in (1).
(1) Who did Pekka see ___?
The wh-element who is extracted from the object position to the left edge. In some circumstances it is not just the wh-pronoun that moves, but a phrase that contains the wh-element, as shown in (2).
(2) Towards which city does the Seine run ___?
In this situation the wh-element is said to pied-pipe the hosting phrase. In the example above, the wh-element which city remains in situ, and the hosting phrase towards which city moves. Huhmarniemi (Reference Huhmarniemi2012a) demonstrates that in Finnish, pied-piping requires an additional step: the wh-element must move to the left edge of the hosting phrase (3).Footnote 1
(3)
In this example, the containing host phrase is an adverb clause. The first movement operation inside the adverb phrase is called secondary wh-movement (Heck Reference Heck2004, Reference Heck2008). The containing host phrase is moved to the final scope position after secondary wh-movement. Both movement operations are mandatory for the formation of a normal interrogative sentence. The fact that the wh-element must occupy the left periphery of the pied-piped phrase is known as the edge generalization (Heck Reference Heck2008:88).
Huhmarniemi (Reference Huhmarniemi2012a) shows that in Finnish, secondary wh-movement + pied-piping applies in an across-the-board fashion, thus applying to preposition phrases (PPs), determiner phrases (DPs), adjective phrases (APs), to several adverb phrases (AdvPs) and even to certain nonfinite constructions. This may lead to a ‘roll up’ movement. Example (4) shows a three-stage process. The interrogative DP is first moved to the left periphery of its host DP, which is in turn moved to the left periphery of the containing PP, which is then moved to the left periphery of the containing CP.
(4)
Every step is mandatory for the creation of an interrogative sentence. Example (4) illustrates recursive pied-piping (Heck Reference Heck2008:76). Not all languages involve extensive roll up movement similar to Finnish. For example, the English equivalents of Finnish pied-piping constructions often involve wh-in-situ. The exact reason for this will have to await further research.
When a phrase XP allows pied-piping (A′-movement or base-generation to the left periphery plus pied-piping), I will call it a pied-piping domain. Example (3) shows that the TUA-adjunct constitutes a pied-piping domain, (4) demonstrates that DPs and PPs are pied-piping domains in Finnish.
But why must wh-elements target the intermediate left peripheral positions on their way up to the final scope position? Why do they not move directly into the final position? A compelling hypothesis is to say that the pied-piped host phrases are islands (phrases which do not let wh-elements to escape). Because they are islands, a wh-element that wishes to move up to the final scope position must make a pit-stop at the edge and pied-pipe the rest. However, Huhmarniemi (Reference Huhmarniemi2012a:229–232) and others have demonstrated that there is optional pied-piping: constructions where both extraction out of the phrase and secondary wh-movement plus pied-piping applies.Footnote 2 Some such constructions are exemplified in (5)–(12) (from Huhmarniemi Reference Huhmarniemi2012a, Reference Huhmarniemib), which illustrate the claim for the MA-infinitival, DPs, PPs and certain adverb clauses.Footnote 3
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
Optional pied-piping shows that pied-piping domains need not be islands.Footnote 4 Another hypothesis is that the intermediate steps instantiate some locality, minimality or economy condition, according to which movement operations must proceed in the smallest possible steps. An auxiliary assumption required is that the edges of pied-piping domains (and only those) function as targets for wh-movement. This hypothesis as such fails to explain why these phrases (and not all phrases, or some other imaginable set of phrases) have such an edge position. This is the question I want to discuss in this paper. I argue that the following generalization emerges from Finnish:
(13) Pied-piping domains (Finnish)
XP is a pied-piping domain if and only if XP may be adjoined.
When a phrase can be adjoined, I will say that the phrase is adjoinable. A deep connection between adjunction, pied-piping and A′-movement is therefore implied.
We clarify the two notions which occur in (13), pied-piping domain and adjunction. The following definitions will be used in this article:
(14) Pied-piping domain
XP is a pied-piping domain (by definition) if and only if a wh-interrogative can stop, or can be base-generated, at its left edge and pied-pipe the phrase upstream, all else being equal.
(15) Adjunction
A phrase XP is adjoinable if and only if it can occur in an adjunct position.
The definition of ‘pied-piping domain’ is not standard. It requires that such phrases satisfy the edge generalization for wh-movement and pied-piping. Pied-piping alone or pied-piping together with wh-in situ are not sufficient. I believe that the results presented here might generalize to such cases as well, but this will not be shown here; hence the more narrow definition.
We provide empirical diagnostic tests to distinguish adjuncts and other phrases. The following standard tests will be used here (see Carnie Reference Carnie2008:121–129):
(16) Adjunct tests
a. Iteration Test
Adjuncts can be iterated inside a given projection.
b. Optionality Test
Adjuncts are optional and cannot be selected by another head.
c. Free Ordering Test
Adjuncts can be ordered freely and do not need to be adjacent to any head.
d. Coordination Test
Adjuncts can be coordinated with other adjuncts, and not with complements or specifiers.
e. Thematic Roles Test
Adjuncts do not bear thematic roles.
f. Projectional Inertness Test
Adjuncts do not change the syntactic properties (e.g., its category or type) of the constituent they are daughters of.
Not all tests apply to every possible situation, and not every adjunct has these properties while some complements and specifiers may in fact have some of them, but it is not controversial that the notion of ‘adjunct’ captures an important linguistic property and that the properties listed in (16) more or less converge into the same set of constituents.
It is important to realize that (13) does not claim that only phrases which are adjuncts constitute pied-piping domains. That would be too strong since, for instance, argument DPs constitute pied-piping domains even if they occur in argument positions. The claim is rather that DP constitutes a pied-piping domain (in whatever context it occurs) because it can be adjoined. For instance, while DPs generally occur in argument positions, as is illustrated in (17a), they may also appear in adjunct positions, as in example (17b), where the DP constitutes a temporal adjunct.
(17)
The claim that the two tokens of the same DP in (17) differ in their syntactic status means that the diagnostic adjunct tests apply to the latter but not to the former.
The definition (14) requires that the wh-interrogative is able to stop at its left edge and pied-pipe the phrase upwards. The condition ‘all else being equal’ abstracts away from conditions which may prevent these operations applying. For instance, it is well-known that agreement may block extraction that would be possible otherwise (Boeckx Reference Boeckx2008, Huhmarniemi Reference Huhmarniemi2009). The definition in (14) is meant to be read so that the agreement factor, among other irrelevant factors considered in this article, is ignored. What is required is that the two key operations are at least possible – hence possible ‘all else being equal’.
The notion of an adjunct, and adjunction more generally, has been subject to several controversies, especially since the emergence of minimalism. Perhaps a few words concerning this debate are in order. There is in my mind no doubt that certain phrases behave like adjuncts in terms of the diagnostic tests listed above, while other phrases fail to satisfy the same tests. Thus, adjuncts and adjunction were defined axiomatically in X-bar theory. Minimalism dispenses with X-bar theory in a way that leaves the distinction between specifiers and adjuncts difficult to capture. The present work does not rely on any analysis concerning the relation between specifiers and adjuncts: I have nothing to say about specifiers. A fully formal theory of adjunction is likewise irrelevant. I will only show that pied-piping domains and adjoinability coincide; if the generalization holds, then any theory of pied-piping, wh-movement and adjunction must explain why the correlation holds.
2. ADJOINABILITY AND PIED-PIPING DOMAINS COINCIDE
Next we argue that any phrase is a pied-piping domain if it can be adjoined, i.e. if it is adjoinable. The argument will be constructed by showing that all adjunct phrases possess the two characteristics. Specifically, in every example below, the (a) example demonstrates that the phrase in question can function as an adjunct or that it occurs in an adjunct position, and the (b) example shows that the phrase in question constitutes a pied-piping domain.
(18)
(19)
(20)
(21)
(22)
(23)
(24)
(25)
(26)
The CP behaves differently with respect to the other pied-piping domains. CPs appear to be in some way ‘too heavy’ to allow smooth pied-piping. They can be moved if embedded inside of a DP (e.g., Sitä että keneltä hän tilaa pizzan Pekka pohti ‘that.d that.c who he order pizza Pekka wondered’). The second, more important, difference is that the CP-internal quantifier does not ‘scope out’ of its clause (Huhmarniemi Reference Huhmarniemi2012a:95). When a non-CP wh-phrase moves, the wh-element can take scope over the material it moves over. This is not true of CP-movement, which must reconstruct obligatorily. Thus, an interrogative such as (27) is ungrammatical:
(27)
It is arguable, therefore, that example (26c) involves topicalization of the complement clause and not pied-piping. If so, then CPs constitute an exception to the proposed generalization: they can be adjoined, but do not constitute pied-piping domains. An independent factor may be in play (see Brattico Reference Brattico2012).
From this data it can be concluded that there is a strong tendency for a phrase that allows adjunction to constitute a pied-piping domain. However, some details need to be clarified before we proceed. I take it as self-evident that PPs, APs and AdvPs satisfy the diagnostic tests (16) for adjuncts. Specifically, they can all be iterated, they are optional, they can be ordered freely, they can be coordinated only with other constituents of the same type, they do not have thematic roles and they are inert inside their projection. Perhaps the claim that the temporal DPs in example (17), repeated here as (28), also constitute adjuncts may require a clarifying remark.
(28)
This DP fulfills the adjunct diagnostics, as shown in (29).
(29)
The Iteration Test, Optionality Test and Coordination Test give different results with complement DPs (30), showing that the adjunct DP contrasts with a complement DP.
(30)
Furthermore, the complement in (30) bears the thematic role of Patient (Thematic Role Test) and it changes the syntactic nature of its projection V → v (Projectional Inertness Test). The Free Word Order Test cannot be applied to a finite clause due to the fact that Finnish word order is relatively free. It can, however, be applied to adjective phrases. In Finnish, an adjective phrase must obey the head-final word ordering, as shown in (31). Thus, all the head–complement dependencies must occur in the head-final order, in contrast to finite clauses where the natural word order is head-initial.
(31)
Adjunct phrases such as koko päivän ‘whole day’ may, however, occur at several positions (32a, b). This behavior contrasts with heads and complements (or arguments), which cannot scramble into any other position (32a, b).
(32)
(33)
Temporal DPs are therefore adjuncts. If a phrase is adjoinable, it constitutes a pied-piping domain. Some authors maintain that bare-DP temporal adverbs are actually PPs headed by empty prepositions (see Demirdache & Uribe-Etxebarria Reference Demirdache, Uribe-Etxebarria, Guéron and Lecarme2004 and references therein), while others hold that they are DPs (Larson Reference Larson1985). If it is true that temporal adverb DPs are PPs, then one could maintain that there are no adjoinable DPs. The hypothesis that pied-piping domains are adjoinable predicts that DPs should be adjoinable. There is strong evidence, however, that bare-DP adverbs are not concealed PPs. Note first that when a DP occurs in the complement position of a preposition, its Case in Finnish is determined by the preposition and never by a nonlocal Case assigner. Therefore, if it is true that bare-DP adverbs are concealed PPs, their Case should never change as a function of the matrix clause properties. This prediction is not borne out. First, in Finnish, the matrix negation requires all direct objects to occur in the partitive, but the same is true of the bare-DP adverb as well, see (34a, b). Secondly, the modal verb täytyy ‘must’ requires direct objects to occur in the nominative Case; it will license nominative Case for the bare-DP adverbs as well, see (34c). In sum, bare-DP temporal modifiers behave like bare DPs. They are transparent to Case assignment.
(34)
I will therefore maintain that DPs are adjoinable. An anonymous reviewer objects to the present generalization by pointing out that (for example) because-clauses are adjoinable but they are not pied-piping domains. This argument might be problematic because the relevant clauses are defined by requiring a specific non-wh-element to occur at their left edge. It is possible to argue against the present generalization by saying that DPs with overt demonstratives are adjoinable but they are not pied-piping domains. However, the reason why such DPs do not involve wh-movement to their left edge is because the overt demonstrative fills the edge position and blocks movement. The same is true of clauses headed by because: there is no space for the wh-interrogative, and therefore such clauses do not pied-pipe.
Next we tackle the ‘only if’ part of our empirical generalization in (13). To show that an XP is a pied-piping domain only if it can be adjoined we must show that the phrases listed in the previous section are all and the only pied-piping domains there are in Finnish. That this holds is argued by Huhmarniemi (Reference Huhmarniemi2012a:208–210). An additional argument can be crafted by going through phrases which cannot be adjoined and showing for each that they do not constitute pied-piping domains. I will go through this demonstration here.
Phrases which cannot be adjoined in Finnish include tense phrases (TPs), small verb phrases (vPs), quantifier phrases (QPs), numeral phrases (NumPs) and the three verbal complement infinitivals (VA, MA, A). The claim is easy to show for the the two infinitivals, VA and A. In each (a) example below a wh-interrogative is formed by extracting the wh-pronoun from the embedded infinitivial; these are all grammatical. In each (b) example, the wh-pronoun and the whole infinitival phrase remains in situ; these are all ungrammatical. Finally, each (c) example shows whether pied-piping is possible. It is possible for the MA-infinitival, contrary to the generalization (13), but not for the VA- and A-infinitival. I will return to the MA-infinitival further below.
(35)
(36)
(37)
The VA- and A-infinitivals do not satisfy the adjunct tests. I will show this for the A-infinitival: they cannot be iterated (38a); they are selected by the governing verb and are often not optional (38b-c); their ordering is not free (38d-e); it is not possible coordinate them with adjuncts (38f) and they bear thematic roles assigned by the governing verb.
(38)
The QP and NumP are projections internal to a noun phrase. They are neither islands (39b) nor projections that can pied-pipe out of the DP, see (39c, d). Rather, as shown by Huhmarniemi (Reference Huhmarniemi2012a:149–151), a wh-element moves to the edge of the DP and pied-pipes the whole DP, as shown in (39b).
(39)
Is it possible to form an interrogative by pied-piping the vP? The matter is difficult to assess, given that both the main verb and the subject may have escaped from the vP before operations apply. It is well-known that a phase whose head has escaped cannot move (Takano Reference Takano2000). But is it possible to find an example where both the subject and the verb remain inside the vP? It is possible to keep the subject inside with two auxiliary assumptions. The first assumption will be that if the clause is headed by an expletive, then the subject remains at (Spec,vP) (Holmberg & Nikanne Reference Holmberg, Urpo and Svenonius2002). The second assumption is that the subject will get nominative Case and trigger phi-agreement at this position and thus will not require movement to (Spec,TP) (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky, Freidin, Otero and Zubizarreta2008). Then we can show that if the external subject is a wh-pronoun, vP cannot be pied-piped to the front of the clause, see (40a, b). On the other hand, it is possible to form an interrogative by moving the wh-pronoun out of the vP, as shown in (40c).
(40)
However, here the verb has climbed out of the vP. To force the verb to remain inside, we need an example where it does not show tense alteration. A relevant example can be constructed by using a tensed modal verb which takes a non-finite verbal complement. The combination of the modal and the nonfinite verb constitutes a monoclausal structure, suggesting that the relevant structure is T + vP. The experiment shows that pied-piping is impossible, see (41a, b).Footnote 5
(41)
Another argument against the assumption that vP constitutes a pied-piping domain can be constructed on the basis of vP adverbs. The hypothesis that vPs constitute pied-piping domains predicts that after the object argument inside a vP has been interrogativized and moved to the left edge of the vP, pied-piping of the vP should bring the vP adverbs along. This prediction is not borne out, as is shown by (42b).
(42)
One direct argument for the conclusion that TP is not a pied-piping domain can be drawn from the fact that the VA-infinitival is not a pied-piping domain. The VA-infinitival inflects for tense and contains, therefore, the TP projection (and arguably nothing else). Another argument is based on the fact that if the TP were a pied-piping domain, complementizers could appear at the end of the finite clause after the whole TP has moved. The resulting construction is ungrammatical in Finnish (43).
(43)
Thus, complementizers cannot be stranded in Finnish. To conclude the argument it must be shown that neither the TP nor vP satisfy the adjunct tests. Neither of them can be iterated; both are selected by the governing head such as C, T or V; they change the category of their projection and their order is not free. I know of no situation where a bare TP or bare vP would behave otherwise.
The only counterexample known to me at present to the generalization that a phrase constitutes a pied-piping domain only if it can be adjoined is the MA-infinitival, which functions as a complement but constitutes a pied-piping domain. This violates generalization (12). The evidence, presented in (36c) above, is repeated here as (44).
(44)
Why is this so? The MA-infinitival differs from the A-infinitival and VA-infinitival in that it constitutes an ECM (Exceptional Case Marking) construction. By being an ECM construction we mean that the case of its thematic subject is assigned by an element in the matrix clause, not by an element within its own clause. A well-known experiment to show this is to use the negation test. In Finnish, the negative word e- forces all direct objects it c-commands to be in the partitive. The matrix negation affects the thematic subject of the MA-infinitival, (45c), but not the subjects of the VA-infinitival, (45b), or the A-infinitival, (45a):
(45)
The thematic subject of the MA-infinitival is therefore in a direct object position. Suppose, therefore, that the structure of the MA-infinitival is such that the thematic subject is not contained in the MA-phrase and is therefore directly case-marked by the matrix clause elements:
(46)
The thematic subject Merjan cannot pied-pipe the MA-infinitival: it is not part of it. An element that is not contained in a phrase XP cannot pied-pipe that XP. There is an alternative derivation which yields (44). According to this derivation, the MA-infinitival syömässä leipää ‘eat.MA bread’ is first fronted by a focus-topicalization rule, after which the wh-pronoun follows, see (47). They can move independently if they are independent constituents.
(47)
This analysis is supported by the fact that the MA-constituent can move as a constituent while stranding the thematic subject:
(48)
This operation triggers a focus/topic interpretation for the MA-infinitival. This in effect replicates the first movement operation assumed to underlie (47). It is then easy to see that if the object Merja is interrogativized, it, too, must climb to the left edge. This produces the apparent ‘pied-piping’ construction. The second supporting fact comes from the observation that native speakers prefer the construction (49), where the wh-element is fronted and the MA-infinitival remains in situ.
(49)
If the ‘pied-piping’ alternative is derived by two movement operations, one which is related to topic/focus structure and another which is related to the interrogativization, then the intuition that the ‘pied-piping’ alternative is more marked is explained.
The third argument derives from the following prediction of the present analysis. Suppose, contrary to the analysis argued here, that the object wh-pronoun pied-pipes the MA-infinitival to the left edge of the matrix clause. This hypothesis predicts that if the MA-infinitival contains a subject, a construction containing the wh-object, subject and the MA-infinitival should be dislocated. This prediction is falsified: such construction is totally ungrammatical, as shown in (50).
(50)
In effect, the presence of the thematic subject prevents pied-piping: only the extraction alternative is licit. On the other hand, this data is explained by the ECM analysis, which says that the thematic subject is not part of the MA-clause. If so, the derivation of (50) performs an impossible operation: it puts an element outside phrase XP, the subject, inside that phrase when they both land at the left periphery.
A fourth argument concerns the fact that an analysis which puts the subject of the MA-infinitival and the residuum into different constituents predicts that it should be possible to move one of them further, stranding the other.Footnote 6 This prediction is confirmed, see (51).
(51)
The fifth argument concerns the slight ungrammaticality of the ‘pied-piping alternative’, shown in (47) above, repeated here as (52).
(52)
I judge the sentence to be slightly off, while some speakers judge the sentence to be close to ungrammatical. Why? The derivation in (52) moves two constituents to the left peripheral A′-position: the residuum MA-infinitive and the wh-interrogative. It is well-known, however, that Finnish sentences have only one left peripheral A′-position (Vainikka Reference Vainikka1989, Vilkuna Reference Vilkuna and Kiss1995, Huhmarniemi Reference Huhmarniemi2012a). This is shown by (53), where a PP autossa ‘inside a car’ is topicalized and wh-pronoun is fronted. I judge the sentence to be quite ill-formed.
(53)
Thus, the question why the pied-piping alternative is ungrammatical can be answered if we assume that it involves two movement operations, both of which target one position.
All in all, it is reasonable to assume that the MA-infinitival contains two parts: the thematic subject, which lies outside the constituent, and the residuum, containing the infinitival and its object. If the thematic subject is outside the MA-adjunct, then we focus on the behavior of the constituent containing the infinitival and its object. Let's call it the ‘residuum MA-infinitival’. Can the residuum MA-infinitival be pied-piped and adjoined? It turns out that both questions are difficult to answer. Example (54) provides a pied-piping construction.
(54)
The bracketed phrase in (54) constitutes the residuum MA-infinitival. To me, it pied-pipes marginally, while the extraction construction is more natural. Is it adjoinable? An example such as (55) suggests that it is.Footnote 7
(55)
The MA-infinitival is optional and, since we have seen that its thematic subject occurs in the matrix clause, it could be argued that it does constitute an adjunct in (55). If so, then the residuum MA-infinitival would be adjoinable, which then explains at once why (54) is possible. This verifies the generalization that adjoinability and pied-piping domains coincide.
However, as much as I think that this would be a desirable result, the validity of the claim that the MA-infinitival is adjoinable is not trivial to establish and this deserves a comment. For one, the MA-infinitival is strongly licensed by certain kinds of verbs, suggesting that it would be subject to s-selection or theta-marking. Adjuncts tend to be immune to selection.
(56)
Second, as noted earlier, the MA-infinitive is not an extraction island. Third, only one or two MA-infinitivals can occur per verb; the construction cannot be iterated. Fourth, the ‘pied-piped’ version seems awkward and more marginal to me than the extraction alternative. This and other reasons have prompted some scholars to regard the MA-infinitive as a complement (Vainikka Reference Vainikka1989, Toivonen Reference Toivonen1995, Manninen Reference Manninen1999) and not as an adjunct. Due to the fact that the residuum MA-infinitival bears some properties of adjuncts and some properties of arguments, it is not possible to evaluate its standing in relation to the present generalization without further exploring the theory of adjunction. My guess is that residuum MA-infinitivals are indeed a mixed case.
This concludes the argument that a phrase is a pied-piping domain if and only if it is adjoinable. Wh-movement, pied-piping and adjoinability are connected with each other at some deeper level of grammatical analysis.
ACKNOWLEDGEMENTS
This project started as a follow-up study to Huhmarniemi (Reference Huhmarniemi2012a). It would not exist without her groundbreaking observations concerning Finnish wh-movement. Arthur Stepanov and Saara Huhmarniemi provided insightful comments which lead to major improvements. I also wish to thank the three anonymous reviewers who provided valuable suggestions and thoughtful empirical and theoretical criticism. This article is the first part of a larger work concerning wh-movement, adjunction and pied-piping. The second part can be found from Brattico (Reference Brattico2012).
