4.5 Extension: Remnant movement and Freezing 77
78 A syntactic procedure for displacement 4 (4.33) a. [x. . . 1. . .]2. . .[. . . a1. . .[. . . 2. . .]]
b. a2. . .[[x. . . 2. . .]1. . . 1. . .]
(4.33a) is a case of remnant movement, and (4.33b) is a Freezing configura-tion. What differs is the derivational order of the two displacements involved.
• Remnant movement
Firstais extracted fromx, then (the rest of)xis displaced.
• Freezing configuration
Firstxis displaced, thenais extracted fromx.
Freezing configurations, i.e. extraction from a displaced phrase, is generally taken to be impossible in languages, whereas remnant movement is assumed to be possible. To illustrate this, consider the following examples from German.
(4.34a) instantiates remnant movement: the NPdas Buchscrambles out of the VP, then the remnant VP is topicalized. (4.34b) on the other hand shows a Freezing effect: first the NP ein Buch wor¨uberis scrambled, and subsequently the PPwor¨uberis extracted from that NP.
(4.34) a. [VP 1Gelesen]2 hat[NPdas Buch]1 keiner 2
read has the book.accno-one.nom
b. ∗[PPWor¨uber]2 hat [NPein Buch 2]1 keiner 1gelesen?
about what has a book.acc no-one.nom read Let us look at how these sentences would be generated with the syntactic mechanism of this chapter. First consider the Freezing configuration (4.34b).
The NP corresponds to the following complex expression (ignoring the semantic dimension):
hwor¨uberwh,(ein Buch:: NP, . . .)i
When this NP is merged with the verb, it would have to be split in order to be displaced itself. But since it is a complex expression, splitcannot apply. This seems desirable because (4.34b) is indeed excluded. But for the same reason also remnant movement is impossible. Consider (4.34a). The VP corresponds to the following expression (where I use Σ as the feature triggering scrambling):
hdas BuchΣ,(gelesen:: VP, . . .)i
This is a complex expression that cannot be split, although it would need to be split in order to be fronted.
Since we do not want to exclude the latter case, the definition ofsplithas to be extended to complex expressions. An obvious way to do so would be to split the nucleus while keeping the edge:
splitha, xi=ha,splitxi
4.5 Extension: Remnant movement and Freezing 79 This indeed allows for remnant movement. The derivation for (4.34a), that got stuck when having to split the expression hdas BuchΣ,(gelesen:: VP, . . .)i, could now proceed. The result of applying splitis:
hdas BuchΣ,hgelesentop,(:: VP, . . .)ii
The derivation proceeds without further complications, with the two forms at the edge being remerged when they can check their features.
But in the same way, the derivation of (4.34b) converges and Freezing effects are no longer obtained. The expression hwor¨uberwh,(ein Buch:: NP, . . .)i can now be split, resulting in:
hwor¨uberwh,hein BuchΣ,(gelesen, . . .)i
Again, the derivation proceeds without further complications and the two forms at the edge are remerged when their features can be checked.
So, the derivations involving Freezing configurations proceed in exactly the same way derivations involving remnant movement do. The reason is the fol-lowing: We decided to split a complex expressionha, xisuch that the result is ha,hx, ii. Above we observed that the difference between Freezing and rem-nant movement is the order in whichaandxare extracted, most importantly in which order aandxare remerged. However, this is a difference we cannot capture. Keeping bothaandxas separate elements at the edge does not give us a way to tell in which order the two will be remerged, and especially it does not give us a way to allow one order and disallow the other.
What to do about it? We need to distinguish the case ofabeing remerged beforex(remnant movement) from the case ofxbeing remerged first (Freezing).
This is actually possible without much ado. The trick is to not splitha, xisuch thataandxare kept as separate forms at the edge but instead such thatha, xi is kept as what it is: one constituent, a complex expression. That is, split should result in the expression hha, xi, i. Once we adapt remerge in order to reach ain this configuration, we are done. What does the job of allowing remerge of a before x (remnant movement) but not vice versa (Freezing) is the fact that remerge is defined only for simple forms. Here is why. If we first remerge a, this is unproblematic because it is a simple form. Then x remains at the edge; if it is a simple form as well, it can also be remerged without a problem. If we, however, tried to first remerge x, it still amounts to the complex form ha, xifor which remergeis not defined. This possibility is therefore blocked.
In a nutshell, we exploit the derivational difference between remnant move-ment and Freezing configurations by extracting ha, xias one constituent and therefore causing a difference in the order of remerging aandx.
Let us look at how this works for our examples above. Recall that the complex expressions we needed to split were (4.35a) in the case of Freezing and (4.35b) for remnant movement.
80 A syntactic procedure for displacement 4 (4.35) a. hwor¨uberwh,(ein Buch:: NP, . . .)i
b. hdas BuchΣ,(gelesen:: VP, . . .)i
Applying split in the way just described now yields (4.36a) and (4.36b), re-spectively.
(4.36) a. hhwor¨uberwh,ein BuchΣi,(:: NP, . . .)i b. hhdas BuchΣ,gelesentopi,(:: VP, . . .)i
In the first case, the scrambling feature can be checked first, thus the whole complex edge at the edge would need to be remerged. (This is something about the understanding of complex expressions and forms in this thesis: the nucleus is the core. It can have an edge or not. However, there can be no edge without a nucleus. So remerging the nucleus always involves the nucleus and its edge.) This remerging fails because remerge is not defined for complex forms.
Also in the second case, (4.36b), the scrambling feature can be checked first.
This leads todas Buchbeing remerged. Since it is a simple form, this is unprob-lematic. The result is the expression hgelesentop,(das Buch keiner:: VP, . . .)i (the form of the nucleus depends on the exact stage of the derivation, which does not play a role here). At some later point in the derivation, the topi-calization feature can be checked. Since gelesen is also a simple form, it can be remerged as well without any problem. The remnant movement derivation hence is perfectly fine.
There is a nice consequence of this derivational approach. It concerns a restriction on remnant movement that M¨uller [82] formulated as the principle ofUnambiguous Domination.
(4.37) Unambiguous Domination
In a structure. . .[x. . . y . . .]. . .,xandymay not undergo the same kind of movement.
Within M¨uller’s representational approach, it is necessary to define a local domain in which this condition applies, because otherwise sentences like (4.38) would wrongly be predicted to be out, for the embedded CP and the more inclusive NP undergo the same kind of movement.
(4.38) [NP Wessen Frage[CP was1 du magst 1]]2 hat 2dich ge¨argert?
whose question what you like has you annoyed For us, this problem does not arise. The displaced wh-expressionwas1is split, percolated and remerged upon constructing the embedded CP. And since it is remerged already when the CP is finished, the CP corresponds to a simple expression. Thus when the NP is constructed and extracted, it will be a simple expression itself. At no point of the derivation does a problematic configuration arise. In fact, the structure in (4.38) is as depicted in (4.32b) at the beginning of this section – a structure that we could already handle without being able to split complex expressions.
4.5 Extension: Remnant movement and Freezing 81 Now I want to extend our formal definitions so they can capture what we just sketched informally. First of all we need to allow pairs of forms at the edge, in other words, allow forms to be recursive.
Form::=String::Cat [Feat]
| hForm,Formi
The definition ofmergestays like it is. But we have to extend the definition ofsplit. For the simplest case it should look like this:
splitha,(b, E)i=hha, bi,(, E)i
And we want to allow this to be recursive and also work if the edge is already a pair:
splithha, bi,(c, E)i=hhha, bi, ci,(, E)i
The general definition comprising these cases and also taking optionality into account is the following:
splithx,h. . . ,(aF, E)ii=hhx, aFi,h. . . ,(, E)ii orhhx, Fi,h. . . ,(a, E)ii
Since we can now have nested forms at the edge, we need to extend what counts as being at the edge. We say that a formxis at the edge of a complex expressionhy, ziif one of the following three conditions holds:
(i) xis equal toy (ii) xis at the edge ofy (iii) xis at the edge ofz
Clauses (i) and (iii) capture the cases we already encountered. For example (i) reaches whatin hwhat, ziand (iii) reaches who in hwhat,hwho, zii. Up to now (ii) would have been equivalent to (i), because y was always simple. Now (ii) captures the new case of reachingwhatinhhwhat, xi, zi. The definition for the functionedgethus now goes as follows:
82 A syntactic procedure for displacement 4
edge(a, E) =∅
edgehx, yi={x} ∪(edgex)∪(edgey)
We also need to generalize the definition ofremerge, in order to consider all new cases of edges, i.e. to be able to not only remerge the first element of a pair but also the edge of this element. It has to capture, for example, cases of the following forms:
• hhaf, xi, y•fi
• hhhaf, xi, zi, y•fi
• hhx,haf, zii, y•fi
Let x[a] denote an expression x in which a occurs, and let x[ ] stand for x where ais removed, andx[b] forxwhereais replaced byb. Then we can give a general definition of remergealong the following lines:
remergex•f[af] =
(x[ ] +a ifa has no more features x[a] otherwise
For the outermost a ∈(edge x), and where xis as before (see Definition 10 on page 67).
What we keep, most importantly, is the restriction that only simple expressions can be remerged.
This concludes how to generalize the syntactic mechanism in order to in-corporate remnant movement while still capturing Freezing effects. This gen-eralization shows one of the possibilities the syntactic mechanism offers due to assuming a recursive structure with respect to displaced elements.
Since we will not need remnant movement in the further course of the thesis, we will consider this section a digression and for reasons of simplicity stick to the less general mechanism of the previous sections.