Covert partial wh-movement and the nature of derivations

Wh-movement is commonly thought to be caused by a syntactic probing operation, initiated by an interrogative probe on C, which triggers subsequent movement to the specifier of C. In this paper I argue that at least English covert wh-movement cannot be described in these terms. I argue instead that covert movement can target positions other than interrogative C, and that this movement is triggered by the interpretational needs of the wh-phrase itself, rather than the formal needs of interrogative C. Evidence will come from the interaction of English multiple whquestions with intervention effects: I document a pattern of intervention effects that is explained only if English in-situ wh-phrases can be interpreted at LF in non-interrogative intermediate positions.


Introduction
Wh-question formation in English involves at least two steps.First, a structure is formed in which a wh-phrase is produced as the argument of a verb or as an adjunct.Second, this wh-phrase is fronted to the left edge of the sentence.Such movement is commonly thought to be caused by an Agree/Attract operation triggered by an interrogative probe on C (Chomsky 1995;2000;and  This paper investigates the behavior of English multiple wh-questions.In a multiple question, only one wh-phrase is pronounced at the left edge of the clause, with all remaining wh-phrases pronounced in-situ, (2a).Recent research on English multiple questions suggests that the (phonologically) in-situ wh-phrase in such questions undergoes covert wh-movement to a position near the overtly fronted one (Richards 1997;Pesetsky 2000;Nissenbaum 2000;Beck 2006;Cable 2007;2010), (2b). 2 (2) The formation of a multiple question: Upon first glance, it is possible to assume that covert wh-movement is triggered by the same mechanism that triggers overt movement in English.This is indeed what the proposals cited above assume.This is most explicitly spelled out in Richards (1997) and Pesetsky (2000): the wh probe on C probes the structure more than once, until all phrases with wh-features have been found and Agreed with.The following step-by-step derivation is assumed for (2): (3) The Agree/Attract model of covert wh-movement: a.
Step 1: the interrogative probe on C probes its c-command domain.
The wh-phrase base-generated higher, which student, is found.
[ CP C +wh [ TP did Fred introduce Which student to which professor?]] b.
Step 2: Which student is attracted to interrogative Spec,CP.
[ CP Which stu.C +wh [ TP did Fred introduce to which prof.?]] c.
Step 3: the interrogative probe again probes its c-command domain.
The wh-phrase base-generated lower, which professor, is found.
[ CP Which student C +wh [ TP did Fred introduce to which professor?]] d.
Step 4: Which professor is attracted to Spec,CP.
[ CP Which stu.which prof.C +wh [ TP Fred introduced to ]] However, upon closer inspection of the behavior of covert wh-movement in English, I will argue that the picture in (3) is untenable.In particular, I will show that covert whmovement can target positions other than interrogative C, in a pattern similar to one that has been argued for recently for overt multiple fronting languages such as Romanian in Cheng & Demirdache (2010) (based on data from Ratiu 2005;2007).I propose that this pattern is best explained if movement is triggered for the interpretational needs of the wh-phrase itself and not by an Attract operation triggered following Agreement between the interrogative probe on C and the wh.The evidence will come from superiority-obeying multiple questions in which a (phonologically) in-situ wh-phrase occurs inside a syntactic island and their behavior with regard to intervention effects.Intervention effects can be used to diagnose whether a wh-phrase is interpreted in-situ in a structure or through covert movement: we observe an intervention effect when wh is LF-in-situ, but not when wh moves (Beck 2006;Kotek & Erlewine in press a.o.).I show that intervention happens when an intervener occurs above an island but not inside it, a pattern consistent with covert movement inside the island to positions other than interrogative C, which cannot escape the island.This constitutes a new pattern of intervention effects not previously documented in English, thus also contributing to our understanding of the correct characterization of intervention effects more generally.
The remainder of the paper is structured as follows: Section 2 provides a brief background on intervention effects.Section 3 focuses on intervention effects in English multiple questions with syntactic islands, and shows that intervention effects occur above the island but not inside it.Section 4 argues that this data motivates a view of wh-in-situ involving covert movement to positions other than interrogative C, and proposes that this movement is best explained as driven by the interpretational needs of wh and not by the traditional feature-driven probing mechanism.Section 5 is the conclusion.

Background: Intervention effects in English questions
The term intervention effect describes a situation in which a question is rendered ungrammatical because an in-situ wh-phrase is c-commanded by an offending intervenercertain quantificational and negative elements, as well as focus-sensitive items -at LF. 5The effects of intervention are best observed in wh-in-situ languages such as Japanese.Although question formation in these languages generally does not require wh-fronting, in-situ wh-phrases must scramble above interveners in order to avoid intervention effects.Example (4) shows this effect for Japanese: (4a) is a canonical question, with a nonintervener subject and an in-situ wh-object.In (4b), this subject is replaced with the NPI intervener 'anyone' without any other change to the question, and the result is ungrammaticality, described as an intervention effect. 6(4c) shows that this structure can be rescued by scrambling the wh above the intervener, resulting in the desired interpretation.what-acc anyone read-neg-past-Q 'What did no one read?'A parallel effect can be observed in German, for wh-in-situ in multiple wh-questions.Example (5a) provides a baseline for a multiple question in German.Here, a non-intervener object 'Luise' c-commands the in-situ wh 'where.' (5b) shows that if this non-intervener is replaced with the intervener 'no one' without any other change to the structure of the question, the result is again an intervention effect.This ungrammaticality can be avoided by scrambling the in-situ wh, so that it is no longer c-commanded by the intervener, (5c).
(5) German: Data such as the above lead to the following descriptive characterization of the intervention effect: an intervener cannot c-command a wh-phrase at LF (6a); to yield an interpretable structure, wh must move above the intervener (6b In English, a more complex pattern is observed.Pesetsky (2000) shows that intervention effects correlate with superiority: intervention is observed in superiority-violating questions but not in superiority-obeying ones.Importantly, at least for some speakers, the intervention effect may only affect the pair-list reading of the question, with a single-pair reading remaining available.The syntax/semantics literature proposes several ways of interpreting in-situ wh-phrases that have not undergone covert movement.These include Alternative Semantics (Hamblin 1973;Rooth 1985;1992), binding (Pesetsky 1987;Tsai 1994), and choice functions (Chierchia 1993;Dayal 1996;Reinhart 1998;Cheng & Demirdache 2010).Of these, only the Alternative Semantics approach has been used in current theories of intervention effects that account for the English data discussed in this paper (see Beck 2006;Kotek 2014).10However, in this paper I will leave aside the details of the semantic theory that explains intervention effects and instead concentrate on a distribution of these effects themselves.

Intervention effects in multiple questions with islands
Given this state of affairs described in section 2, we can use intervention effects as a diagnostic for whether or not covert wh-movement has taken place in the derivation of a question: the presence of an intervention effect teaches us that a (phonologically) in-situ wh-phrase must be interpreted in-situ, below the intervener, whereas the lack of an intervention effect teaches us that the wh-phrase must have covertly moved above the scope of the intervener.See also Kotek & Erlewine (in press) and Erlewine & Kotek (2014) for other arguments motivating this diagnostic.This will be the goal of section 3 of the paper.
The crucial data for this paper will come from the interaction of intervention effects with English multiple questions in which the (phonologically) in-situ wh-phrase occurs inside a syntactic island.I take as my starting point the fact that multiple questions with islands can have pair-list readings, as illustrated in example (11) from Cheng & Demirdache (2010), who attribute the context in (11) to Chris Tancredi (p.c.).This is contrary to a claim by Dayal (2002) that such questions do not have pair-list readings.11(11) Multiple question with island can also have a pair-list reading Context: Each of two philosophers will be offended if we invite one of two linguists.What I want to know is: Which philosopher will be offended if we invite which linguist?a.  Pair-list: Quine will be offended if we invite Chomsky, and Lewis will be offended if we invite Kayne.b. #/* Single pair (infelicitous due to context): Quine will be offended if we invite Chomsky.
This is important since it is specifically the pair-list reading of a multiple question that is sensitive to intervention effects.With a single exception, I have found that native speakers of English find the target sentence in (11a) grammatical and felicitous in the context.With this baseline established, we now turn to an investigation of the presence and extent of covert wh-movement in English multiple questions with islands.
As shown in section 2, superiority-obeying multiple questions in English are generally exempt from intervention effects.Focusing on questions with islands, we will ask two related questions about covert wh-movement.First, if covert wh-movement happens, must it target interrogative C? This is standardly assumed in theories of interrogative semantics: the mechanism that interprets questions through movement requires all wh-phrases to occupy positions local to interrogative C for them to be interpretable (Karttunen 1977 and much subsequent work).Alternatively, wh may be interpreted in-situ without movement (Hamblin 1973 and much subsequent work).This all-or-nothing stance is often implicitly adopted in theories of interrogative syntax, including those which have been proposed for English multiple questions, cited above.A third option, compatible with at least some in-situ approaches to wh-in-situ, is to allow wh to be interpreted in positions that are not interrogative C nor fully in situ.This will be the view endorsed in this paper.
Second, is covert wh-movement sensitive to syntactic islands?If movement is able to target positions other than interrogative C -as I will argue below -one way to show this is to restrict its possible landing sites.Syntactic islands are known to block overt movement out of them (Ross 1967), and they have been argued to affect at least some instances of covert movement (e.g.Huang 1982).If covert movement in English differs from overt movement only in the choice of which copy of the movement chain is pronounced (as in e.g.Bobaljik 1995;Chomsky 1995;2000;or Pesetsky 2000), it should be similarly sensitive to islands.
With this background in mind, let us turn our attention to the data.To preview, I will show that the following generalization holds: (12) Generalization: the interaction of intervention effects and islands Intervention occurs when an intervener is placed above an island containing a wh, but not when an intervener is placed inside the island.
First, let us re-examine example (11), repeated here as the slightly modified (13).12As shown above, this question has two felicitous readings: a single-pair and a pair-list reading.Since in this section we are only interested in the presence or absence of the pair-list reading of a given question, I will restrict my attention to this reading alone.All the examples below have felicitous single-pair readings. 1313) Lower wh inside adjunct island: pair-list reading is available Context: The linguists at the conference are very picky about attending the conference dinner.However, each of them adores one philosopher and will certainly attend the dinner if that philosopher is invited.What I want to know is: Q: Which linguist will come [if we invite which philosopher]?A: Chomsky will come if we invite Quine, Kayne will come if we invite Lewis, Labov will come if we invite Russell, ...
Similarly, when the in-situ wh-phrase is inside a Complex NP (CNP) island, the resulting question can have both a single-pair reading and a pair-list reading. 14(14) Lower wh inside CNP island: pair-list reading is available Context: The linguists at the conference are very suspicious of rumors.However, each of them believed one of the rumors going around that we invited a particular famous philosopher to the conference party.What I want to know is: Q: Which linguist believed the rumor [that we invited which philosopher]?A: Chomsky believed the rumor that we invited Quine, Kayne believed the rumor that we invited Lewis, Labov believed the rumor that we invited Russell, ... Examples ( 13) and ( 14) thus provide us with baselines for the crucial test cases.Next, we introduce interveners into these questions, ( 15)-( 16).We find that an intervention effect, diagnosed by the loss of the pair-list reading, occurs when an intervener (here: only or negation, in bold) occurs above the island, but not when it is inside the island.Furthermore, configurations with three wh-phrases similar to those studied by Cheng & Demirdache (2010), where two wh-phrases are inside an island and one is outside, again exhibit intervention effects, diagnosed here by the loss of the list of triples reading, when an intervener occurs above the island (18a) but not when it is inside it (18b).This is again predicted by the generalization in ( 12) and is consistent with the behavior of multiple questions that we have seen in ( 15)-( 17).
(18) Questions with three wh: intervention above but not inside the island a. Q: Which linguist didn't believe the rumor [that which student invited which philosopher]?* PL A: Chomsky didn't believe the rumor that Mary invited Quine, Kayne didn't believe the rumor that Jane invited Lewis, Labov didn't believe the rumor that Anya invited Russell, ... b.Q: Which linguist believed the rumor [that which student didn't invite which philosopher]?A: Chomsky believed the rumor that Mary didn't invite Quine, Kayne believed the rumor that Jane didn't invite Lewis, Labov believed the rumor that Anya didn't invite Russell, ...
Note that if two wh-phrases occur outside the island and only one wh-phrase is inside it, we predict a pair-list reading with the answer for the third wh held constant.This prediction is borne out, as illustrated by the possible answer in (19a), where only the higher pair of wh-phrases vary and the third wh is held constant, as opposed to the unavailable answer in (19b), where all three wh-phrases vary simultaneously.Finally, I note that if the in-situ wh-phrase can be given exceptionally wide scope, so that it occupies a position above the intervener at LF, we expect the question to become grammatical again.Here I will use extraposition, building on Williams' generalization and the extension of this logic in Fox & Nissenbaum (1999): (20) Williams' generalization (Williams 1974: ch. 4): When an adjunct β is extraposed from a "source NP" α, the scope of α is at least as high as the attachment site of β (the extraposition site).
Given Williams' generalization, we expect extraposition of a relative clause containing an in-situ wh in examples like ( 16) to assign the wh wider scope than when it is not extraposed.The relevant example is given in (21) below.Example (21a), repeated from (16a), provides a baseline without extraposition, where the pair-list reading is blocked because of an intervention effect. 19When the noun-complement clause is extraposed above the temporal adjunct yesterday, giving the clause exceptionally wide scope above the intervener, the pair-list reading then becomes available (21b). 20 (21) Extraposition allows exceptional wide scope for in-situ wh: a. * PL Which philosopher didn't believe the rumor that we invited which linguist?b. ?Which philosopher didn't believe the rumor yesterday [that we invited which linguist]?
To summarize, the structural description of the configuration yielding intervention effects can be summarized as in ( 22), repeated from above: (22) Generalization: the interaction of intervention effects and islands Intervention occurs when an intervener is placed above an island containing a wh, but not when an intervener is placed inside the island.
The data presented here instantiates a previously undescribed pattern of intervention effects in English superiority-obeying questions.This constitutes a counter-example to the generalization in Pesetsky (2000) and subsequent work that intervention effects in English correlate strictly with superiority, such that superiority-violating questions are sensitive to intervention effects while superiority-obeying questions are immune from such effects.Moreover, this finding has important implications for theories of interrogative syntax/semantics, which I explore in the next section.

Covert movement and the nature of syntactic derivations
I began section 3 by asking two related questions.First, if covert wh-movement happens, must it target interrogative C? Second, is covert wh-movement sensitive to syntactic islands?In this section I will argue that the intervention pattern in (22) shows that covert movement must take place in the English questions discussed, that this movement is sensitive to syntactic islands, and that it must be able to target positions other than C. As discussed in section 2, current theories of interrogative syntax/semantics assume that a wh-phrase must either covertly move to interrogative C or else be interpreted in-situ at LF. 21 For example, following Pesetsky (2000) and others, we may assume that superiorityobeying questions with interveners such as (23a) have derivations such as (23b).Here, the (phonologically) in-situ wh-phrase which philosopher covertly moves to C at LF, correctly predicting the lack of an intervention effect in that question.( 23 Next consider examples (24a-c), summarizing the pattern discovered in section 3.If we assume that covert movement is insensitive to islands, movement of which philosopher to C -along the lines of the derivation sketched in (23b) -should be possible for all of (24a-c).This clearly cannot derive the correct judgment pattern, and in particular the ungrammaticality of (24c).Hence, we conclude that covert movement is sensitive to syntactic islands.
( This conclusion has consequences for our understanding of the nature of islands.First, the data in this paper is unexplained if islands are a PF phenomenon, as often assumed for example in the literature on the amelioration of island effects through ellipsis (cf Ross 1969; Merchant 2001, among many others).Under such an approach, islands should not restrict covert movement, and we would hence expect no intervention effects in any of the questions in (24a-c), contrary to fact.Moreover, a theory in which islands can be covertly pied-piped to C, as has been proposed for Japanese (Nishigauchi 1990;Richards 2000), incorrectly predicts no intervention effects for interveners occurring above the island.Under such a theory, the entire island in (24c) would undergo massive pied-piping to the matrix interrogative Spec,CP, resulting in a structure in which wh is not c-commanded by the intervener and hence no intervention effect would be expected. 22lternatively, we might imagine that in-situ wh-phrases never undergo covert movementthat is, an approach in which which philosopher is interpreted in-situ at LF.This type of derivation would correctly predict that (24a) is grammatical and that (24c) is ungrammatical, since it contains an in-situ wh-phrase c-commanded by an intervener.However, this type of derivation would incorrectly predict that (24b) should also be ungrammatical, because here too we would have an in-situ wh-phrase c-commanded by an intervener.Hence, a derivation without any covert wh-movement is untenable.
Instead, the derivation that would predict the full pattern in ( 24) is one involving a partial movement step to a position above the intervener and possibly as high as the edge of the island, followed by in-situ interpretation of the wh between the landing site of movement and C.This partial movement proposal lends support for a theory of interrogative syntax/ semantics that involves both movement and in-situ interpretation within the derivation of a single wh-question (cf Pesetsky 2000;Beck 2006 for superiority; Cable 2010; Kotek & Erlewine in press for pied piping). 23However, the two mechanisms are used here in a novel order of operations: first, a short movement step occurs.Then, the wh is interpreted in-situ between this landing site and C. See Kotek (2014) for a semantic proposal that is able to interpret structures with the syntax that I am proposing here.Two possible characterizations of partial movement are summarized in ( 25): (25) Two ways to characterize partial movement: a. Move as much as possible: Move as close to interrogative C as possible in the derivation.
In the absence of islands, move all the way to C. In the presence of an island, move to the edge of the island.b.Move just as much as necessary: Move only as far as necessary to render a structure interpretable.All things being equal, remain in-situ.In the presence of an intervener, move above it but no further.
Both characterizations of partial movement are able to derive the pattern of judgments in (24a-c), but they have different consequences.If movement targets a position as close to C as possible, the presence of an intervener is irrelevant as a trigger for movement.Lacking an island, we predict wh-movement to always target interrogative C. If an island is present, movement would always target its edge.We would hence predict the derivation in (26a) for the question in (24a), which lacks an island.We predict the derivation in (26b) for the question in (24b), which has an intervener inside an island.And we predict ungrammaticality in the case of (24c), since an intervener occurs above the island, but movement can only target the edge of the island but no further.

ISLAND
Alternatively, if covert movement takes place just when necessary and targets the first position that would yield an interpretable structure, interveners act as triggers for movement.In the absence of an intervener, all things being equal, we expect no movement to take place.If an intervener is present, movement would target a position immediately above the intervener, unless this movement is blocked by an island barrier.
We would hence predict a derivation without any movement for the question in (24a), which lacks an intervener (27a).We predict a derivation for the question in (24b) with wh moving just above the intervener inside the island, (27b).And we predict ungrammaticality in the case of (24c), since an intervener occurs above the island, but movement cannot escape the island.

ISLAND
These two possible characterizations of partial movement follow naturally from two different models of structure building in syntax: I will show here that a bottom-up model would predict movement as in the 'move as much as possible' view, while a left-to-right, top-down model would predict movement as in the 'move just as much as necessary' view.I discuss both of these options in more detail below.

Bottom-up derivations: English as covert Romanian
Consider first a bottom-up model of syntax.In such a model, wh-phrases carrying interrogative features enter into the derivation in earlier cycles (phases) than the C that they eventually Agree with.24Wh-phrases are then 'carried along' into higher phases using edge features (Chomsky 2001;2008, among others) or a similar mechanism, or through Greedy movement (Chomsky 1995).In a question without islands, movement is thus predicted to terminate in interrogative Spec,CP, where wh and C can enter into a local relationship.This view of covert wh-movement is standardly assumed in work on English questions, for example in Richards (1997) and Pesetsky (2000), who propose to view English as a covert version of a multiple wh-fronting language, such as Bulgarian or Romanian.Bulgarian and Romanian require all wh-phrases in a question to overtly front to the specifiers of interrogative C. English question LFs are argued to have a similar structure, but there is a pronunciation rule that dictates that only one wh-phrase is pronounced in its moved position, with all other wh-phrases pronounced in their base-generated positions. 25 If a barrier to movement such as an island is present, we may assume that it will restrict movement, so that the wh-phrase will move to the edge of the island but be unable to proceed any further.This will give rise to a structure as in (26b), and correctly predict the observed generalization regarding intervention effects in English multiple questions, ( 22).
As noted by Cheng & Demirdache (2010), citing Ratiu (2005;2007), a derivation of this form is overtly exemplified in multiple wh-questions in Romanian.Romanian is a multiple wh-fronting language, normally requiring all wh-phrases in a multiple question to overtly front to specifiers of interrogative C.However, in the presence of an island we observe a different behavior: wh cannot move outside an island, (28a), and it also cannot stay in its base-generated position, (28b).Instead wh moves to the edge of the island, (28c). (28) Overt multiple fronting in Romanian questions (Ratiu 2005;2007)  Cheng & Demirdache (2010) motivate a partial movement derivation for English questions with islands from this parallel with the overt behavior of Romanian, and in addition from a consideration of the readings of English questions with three wh-phrases where some wh-phrases are "trapped" inside an island.The proposal in ( 26) is thus equivalent to Cheng and Demirdache's proposal.See Cheng & Demirdache (2010) for details.
This model of structure building is thus naturally consistent with a move as much as possible approach: in most cases, we will end up with a derivation in which covert movement targets interrogative Spec,CP, as in the traditional theory (Karttunen 1977).However, within this model, movement must be caused by the needs of the wh itself and not by the needs of C: if C required all wh-phrases to reach its edge, be it for syntactic or for semantic reasons, we would be unable to correctly predict any partial movement and hence be unable to model the pattern of intervention effects presented in section 3 of the paper.
However, once a syntax and semantics are put in place that allow wh-phrases to be interpreted without requiring movement to interrogative C, the mechanisms of edge features and Greedy movement become conceptually more difficult to justify.
Edge features may fail to Attract a wh across an island but not lead to a crash in the derivation (Preminger 2011).Similarly, the Greedy movement mechanism pushes a wh-phrase as far as it can, but the derivation does not crash if the wh does not reach its destination.But why would we assume the existence of these mechanisms in the first place, if we independently must assume a syntax/semantics that is able to interpret wh-phrases partially or completely in-situ, in non-interrogative positions? 26 more parsimonious syntax would not assume such syntactically-driven mechanisms as a general rule, and instead would model covert wh-movement as triggered only when necessary for the interpretational needs of the question: here, wh cannot be interpreted if it is c-commanded by an intervener and hence the only convergent LF for such a structure must involve movement of the wh above the intervener.Out of considerations of Economy, such movement would, all things being equal, be predicted to be as short as possible, targeting a position immediately above the intervener.This, however, is not straightforwardly accommodated in a bottom-up model of syntax, involving probes and goals, edge features, or Greedy movement, without encountering a lookahead problem: how can we know what is the lowest position at which the wh is interpretable before we know what the path between wh and C looks like -and specifically, where interveners occur, if any?
We may be able to predict that movement will target the lowest possible position in a bottom-up model through a theory of Economy that utilizes trans-derivational competition (e.g.Reinhart 2006), but such approaches have recently fallen from grace. 27,28 However, this movement is naturally predicted in a top-down model of syntax that does not require trans-derivational competition, as I will illustrate next.

Left-to-right, top-down derivations: English as covert German
Consider next a left-to-right, top-down model of syntax, often used to describe how parsing of natural language takes place (Phillips 1996, among others).Under such a model, the parsing of a question is straightforward in English: the presence of an interrogative C heading a question is made explicit by the overt fronting of a wh-phrase to Spec,CP and T-to-C movement.It is now possible to keep track of any interveners and islands occurring inside this question.If a second wh-phrase is encountered -marking the fact that we are constructing a multiple question -it is immediately clear that this wh-phrase cannot occupy a position below an intervener, because the wh would not be interpretable in such a position.The lowest target position of covert movement is also immediately clear: it is the position immediately above the intervener. 29If the wh is trapped inside an island, the extent of movement will be clearly limited by its presence, which would also have already been encountered before the wh-phrase is encountered.
For illustration purposes, it may be beneficial to think of the covert movement envisioned by this kind of approach as movement to the right, as in the (simplified) sketch below: 30 27 Such approaches have been argued to be computationally intractable, too powerful, and to make incorrect empirical predictions.See in particular Collins (1996), Sternefeld (1996), Johnson & Lappin (1999), Frampton & Gutman (1999), Potts (2001), andGärtner (2002).Compare also Chomsky (1995;2000;2001) to earlier Minimalist work in Chomsky (1992;1993), as discussed in Reinhart (2006: Ch.1).See Graf (2013) for recent arguments against these objections. 28For another approach which may derive partial movement without the need for transderivational constraints, see Heck & Mueller (2000). 29A longer movement step could be possible but may be ruled out by considerations of Economy (Chomsky 1995;Fox 1995;2000). 30However, nothing hinges on this particular approach.But see Fox & Nissenbaum (1999) for an empirical argument for covert movement being rightward in this way.
The (phonologically) in-situ wh-phrase which philosopher cannot be integrated into the structure in-situ, as this would lead to an intervention effect because of the c-commanding sentential negation.As soon as this wh-phrase is encountered, the structure must undergo a reanalysis step, resulting in the integration of the wh in a position above the intervener.This reanalysis step is parallel to the reanalysis required in cases of garden-path sentences: (30) A garden path sentence I convinced her children are noisy.
Here, the structure undergoes reanalysis from an initial assumption by the parser that a DP her children should be constructed to a structure in which her alone acts as the first internal argument of convince, and children are noisy acts as the second argument. 31his model of structure building is thus naturally consistent with the move just as much as necessary approach, where movement happens only when it is necessary for interpretability.Therefore no movement happens if no intervener is present in a question, and movement would target a position immediately above an intervener if one is present, to avoid an intervention effect.However, movement cannot escape an island, and hence interveners occurring outside the island lead to ungrammaticality.
This state of affairs is parallel to cases of overt scrambling observed in German multiple questions.As mentioned in section 2, in-situ wh-phrases in German scramble above any interveners in the structure in order to avoid ungrammaticality.The relevant data is repeated here from (5) above: although in-situ wh-phrases can normally be interpreted in their base-generated position (31a), they cannot be c-commanded by an intervener and instead must scramble above it (31b-c).( 31 We can capitalize on this parallel by proposing that covert wh-movement in English is a more restricted, local operation than the unbounded dependency we normally conceive of for overt wh-movement.Specifically, it may be fruitful to think of covert whmovement as a form of covert scrambling.If proposals such as Johnson & Tomioka (1997) are on the right track, Quantifier Raising in English should also be recast as covert scrambling.If so, English has exactly one covert scope-taking operation -scramblingparallel to its overt counterpart in German.English, then, is not a covert version of Romanian, as often assumed in the literature (e.g.Richards 1997), but instead a covert version of German. 32uch a proposal may be advantageous from an acquisition point of view, as well as from general principles of parsimony.Proposals that assume long-distance covert movement alongside the more restricted QR in English may face more difficulty than theories that implement just one type of covert movement operation.Moreover, we can explain how the overt vs. covert nature of scrambling can be acquired by a child: there will be ample evidence in the input that German allows overt scrambling.If the lack of evidence for an overt variant of the operation signals to the child that the target language contains the covert variant of it, the child will acquire covert scrambling in English.
This state of affairs thus leads to the conclusion that overt wh-movement and covert whmovement may be qualitatively different from one another.While overt wh-movement always targets one and the same position -the specifier of interrogative C -and must take place without exception, covert movement may target different positions in the structure and may not occur at all in some cases.

Conclusion
The behavior of English superiority-obeying multiple wh-questions in which the lower wh is trapped inside an island with regard to intervention effects sheds light on the syntax and semantics of wh-questions.I motivate the generalization that intervention effects occur above the island in such cases, but not inside it.This state of affairs requires a syntax that allows for partial covert movement of wh-phrases to intermediate, non-interrogative positions in the derivation, followed by in-situ composition between the landing site of wh and C.This is not compatible with a probe-goal approach to movement triggered by the needs of C, such that C requires the wh-phrase to occupy its specifier, nor is it compatible with semantic theories where wh-phrases must stand in a local relation with C in order to be interpretable, such as the influential Karttunen (1977) analysis and subsequent work.
I discuss two possible characterizations of this covert partial wh-movement: move as much as possible, predicting frequent covert movement to C, and to the edge of an island if one is present; and move just as much as necessary, normally leading to no movement at all, and movement above interveners if they are present, unless there is an island which blocks this movement.I argue that these two characterizations are naturally accommodated in two different approaches to derivation in syntactic theory: bottom-up vs. left-to-right/top-down.Bottom-up derivations, driven by the traditional probe-goal model of syntax, predict intermediate movements through edge features as far as possible, moving all the way to interrogative C unless it is blocked by a barrier along the way.As such, we may conceive of English as a covert version of Romanian.Alternatively, left-to-right, top-down derivations, based on parsing considerations, predict movement to be as short as possible and to be triggered only when necessary, akin to covert scrambling.Under this view, we can conceive of English as a covert version of German.I point out some conceptual difficulties with the bottom-up approach to the derivation of covert wh-movement, favoring instead the top-down approach.Ultimately, however, I do not rule out either approach, and instead leave the resolution of this issue open for future research.
'move as much as possible' view: a.No island: Move to Spec,CP b.Island: Move to edge of island 'move just as much as necessary' view: a.No intervener: No movement b.Intervener: Move above intervener 3,4 a. Which student did Mary introduce to which professor?b.LF: [ cp Which stu.which prof.C + wh [ TP Fred introduced to ]] ).
CP C ... intervener ... wh] b. [ CP C ... wh intervener ... t] 7As a result, from this point on, I will use the notation * PL to indicate this lack of a pair-list reading due to an intervention effect. 8 (Zwicky 1971; at the conference don't really want to attend the conference dinner.However, each of them adores one philosopher and has said that they will come just in case that philosopher is invited.What I want to know is: Q: Which linguist will only come [if we invite which philosopher]?* PL A: Chomsky will only come if we invite Quine, Kayne will only come if we invite Lewis, Labov will only come if we invite Russell, ... b.Context: The linguists at the conference are looking forward to the conference dinner.However, each of them dislikes all but one philosopher and will attend the dinner just in case that philosopher alone is invited.What I want to know is: Chomsky believed the rumor that we didn't invite Quine, Kayne believed the rumor that we didn't invite Lewis, Labov believed the rumor that we didn't invite Russell, ... Some speakers report similar contrasts in questions with an in-situ wh-phrase inside the complement clause of a non-bridge verb, such as dream or shout.Such verbs have been argued to be islands for extraction(Zwicky 1971; Erteschik-Shir 1973, a.o.), and we therefore predict that interveners that occur above the complements of such verbs but not ones 15 Example (15a) is equally ungrammatical under the pair-list reading if only follows come.I thank an anonymous reviewer for bringing this to my attention.occur inside them (17b) should cause an intervention effect, diagnosed by the lack of a pair-list reading.This prediction is indeed borne out, as illustrated in (17a-b).
Q: Which linguist will come [if we only invite which philosopher]?A: Chomsky will come if we only invite Quine, Kayne will come if we only invite Lewis, Labov will come if we only invite Russell, ... (16) CNP island: intervention above but not inside island a. Context: The linguists at the conference are very gullible and believe lots of rumors.However, each of them is suspicious of one rumor about a philosopher that we supposedly invited to the conference party.What I want to know is: Q: Which linguist didn't believe the rumor [that we invited which philosopher]?* PL A: Chomsky didn't believe the rumor that we invited Quine, Kayne didn't believe the rumor that we invited Lewis, Labov didn't believe the rumor that we invited Russell, ... b.Context: The linguists at the conference are very suspicious of rumors.However, each of them believed the rumor that we failed to invite one philosopher to the conference party.What I want to know is: Q: Which linguist believed the rumor [that we didn't invite which philosopher]?A: (i) * PL Which linguist will come only [if we invite which philosopher]? 16A reviewer points out that the question in (15b) degrades if it is embedded under know, as in (ia) below.I agree with this judgment, and note that, in fact, the question is degraded under other embeddings such as want to know (ib), as well.I suspect that this effect is due to properties of the embedding which are independent of intervention effects, but I leave this as an open issue for future research.(i) a. # John knows which linguist will come [if we only invite which philosopher].b. # John wants to know which linguist will come [if we only invite which philosopher].that PL Which protester didn't shout [that we invited which politician]? b.  Which protester shouted [that we didn't invite which politician]?
Chomsky didn't tell Quine about the rumor that Mary had won..., Kayne didn't tell Lewis about the rumor that Mary had won..., Labov didn't tell Russell about the rumor that Mary had won..., ... b.Impossible answer: list of triples (varying all whs at once) Chomsky didn't tell Quine about the rumor that Mary had won..., Kayne didn't tell Lewis about the rumor that Jane had won..., Labov didn't tell Russell about the rumor that Anya had won..., ...
18 (19)Questions with three wh: pair-list reading for whs above the island Which linguist didn't tell which philosopher about the rumor [that which student had won a dissertation prize]?a. Possible answer: list of pairs (varying only the higher pair of whs) ) Covert movement is available: no intervention in superiority-obeying question a.  Which linguist didn't invite which philosopher? b.LF: [ CP Which linguist which philosopher [ C [ TP t 1 didn't invite t 2 ]]] a. Wh can't move out of the island: *[ CP cine i ce k [ IP t i o cunoaşte pe studenta who what cl.3.fsknow prep student [ ISLAND căreia i s-a dedicat t k ieri?]]] which.datcl.dat expl.auxdedicated yesterday b.Wh can't stay in-situ: *[ CP cine i [ IP t i o cunoaşte pe studenta who cl.3.fsknow prep student [ ISLAND căreia i s-a dedicat ce k ieri?]]] which.datcl.dat expl.auxdedicated what yesterday c.Wh moves to the edge of the island: [ CP cine i [ IP t i o cunoaşte pe studenta who cl.3.fsknow prep student [ ISLAND căreia ce k i s-a dedicat t k ieri?]]] which.datwhat cl.dat expl.auxdedicated yesterday 'Who knows the student, to whom what was dedicated yesterday?'