Wat dacht je wat

1
Wat dacht je wat?

• On the strategies for the formation of
  long-distance dependencies in constituent
  questions

Marcel den Dikken (CUNY Graduate Center
KNAW) Hans Bennis (Meertens Instituut, KNAW)
KNAW Visiting Professors Lecture Series 15
October 2009
2
Imagine someone stole your bike the other day,
and you would like to find out who did it.
You can make an enquiry about this by asking
a question of particular kind a question that
places the constituent that represents the thief
in sentence-initial position, in the form of a
question word, as follows
(1) who stole my bike?
The fact that who, the wh-phrase, is in
sentence- initial position here is not very
remarkable its the subject of the sentence, and
in a language such as English, the subject of the
sentence is usually in sentence-initial position
(2) you stole my bike
3
Now imagine you would like to find out how
many other bikes this thief has stolen.
The question you will ask for this purpose again
places the wh-phrase in sentence-initial position
(3) how many other bikes have you stolen?
This time the fact that how many other bikes is
in sentence-initial position is much more
remarkable the direct object in English is not
normally in sentence-initial position
(4) you have stolen a lot of bikes
It seems that in (3) the wh-object has migrated
to the left edge of the sentence.
4
Next, imagine that you heard Hans say that
someone stole his bike, but you dont remember
who.
In such a situation, you can ask the following
kind of question
(5) who did Hans say stole his bike?
This is a long-distance wh-question the
wh-phrase who belongs to the clause embedded
underneath say, but it is pronounced in the
higher clause, once again in sentence-initial
position.
So one thing that is interesting about this
wh-question is that the wh-phrase is not
pronounced in the clause in which it belongs
but instead in a higher clause.
5
Another interesting thing about the long-distance
wh- question in (5) is that the subordinating
conjunction that, which can otherwise introduce
an embedded finite clause in English, must be
absent (5') is bad.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
This suggests that the long-distance wh-question
in (5) is derived by movement of who from the
clause embedded underneath say into the higher
clause such movement cannot proceed across that
if the wh-phrase is a subject that blocks the
movement of who.
(5) who did Hans say (that) who stole his bike?
6
In contrast to English, in Dutch a
long-distance subject wh-question must include
the subordinating conjunction dat while in
English (5) is good and (5') is bad, in Dutch
the situation is exactly the reverse
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
That (6) is bad in Dutch is unremarkable dat is
always obligatory at the left edge of an embedded
finite clause.
7
But the fact that (6') is good Dutch is
remarkable if (5') follows from a ban on
movement of who across that then the
grammaticality of (6') could indicate that in a
Dutch long-distance subject question, there need
not be movement of the wh-phrase out of the lower
clause.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
8
But the fact that (6') is good Dutch is
remarkable if (5') follows from a ban on
movement of who across that then the
grammaticality of (6') could indicate that in a
Dutch long-distance subject question, there need
not be movement of the wh-phrase out of the lower
clause.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
But how can (6') not involve movement of wie out
of the embedded clause? Isnt wie the subject of
the embedded clause in (6') just as who is the
subject of the embedded clause in (5)?
9
Indeed it is but it seems that in the
Dutch-speaking world, wh-phrases that belong to
an embedded clause can originate in a higher
clause, and establish their connection with the
lower clause via an associate in the lower
clause.
Thus, several Dutch dialects can make
long-distance wh-questions by employing two
tokens of the wh-phrase the so-called
wh-copying construction, illustrated in (7)
(7) wie zei Hans wie zn fiets had gestolen?
Here the real wh-phrase is the wie in first
position but its not alone it has an identical
twin at the left edge of the embedded clause.
10
The wh-phrase in the higher clause and its
associate in the lower clause arent always
identical twins there are also dialects of Dutch
in which the associate is a relative pronoun,
die
(8) wie zei Hans die zn fiets had gestolen?
This confirms that the wh-phrase in first
position is indeed the real wh-phrase, but it
casts doubt on the idea that literal wh-copying
might be involved.
To round out the empirical picture, lets add two
more options, both attested in Dutch dialects as
well in both (9) and (10), the wh-element in the
higher clause is the invariant wh-word wat, with
the embedded clause having the same two choices
as before either wie or die.
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
11
So in summary, this is what we find in the
Dutch- speaking world, for long-distance subject
wh-questions
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
It is important to bear in mind that all of these
ways of asking the question of who Hans said
stole my bike mean exactly the same thing but
they have different distributions across the
Dutch-speaking world.
The following slides give you a sense of the
regional distribution of these long-distance
subject wh-questions.
12
map 1 (7)
13
map 2 (9)
14
At this point, lets return to the summary of the
ways of forming long-distance subject
wh-questions in Dutch
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
For collective reference, lets call the examples
in (7)(10) the multiple exposure cases.
15
We should also be aware of the logically possible
multiple exposure cases that do not
materialise
(6) wie zei Hans zn fiets had gestolen?
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(8') die zei Hans wie zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(9') wie zei Hans wat zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
(10') die zei Hans wat zn fiets had gestolen?
16
So heres, once again, whats possible in Dutch
long- distance subject wh-questions
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
How do we make sense of this pattern, and how do
we explain the fact that Dutch differs from a
language like English in (a) allowing (6') and
(b) having the multiple exposure cases in
(7)(10)?
In the remaining time, we will provide a broad
sketch of our answers to these questions.
17
Lets start with the multiple exposure cases,
(7)(10).
For these, we propose that, throughout, the
wh-element in the higher clause is born in that
clause that is, none of these examples involve
long wh-movement.
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(9') wie zei Hans wat zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
For (9), this is perhaps most straightforwardly
the case wat is clearly not the subject of the
embedded clause all it does in this sentence is
mark the scope of the wh-phrase wie, which is
pronounced in the lower clause.
The ungrammaticality of (9') follows wat in the
lower clause cannot mark scope, and is hence
redundant.
18
Concretely, the person, number, and gender
properties (including animacy) of the wh-word in
the lower clause are copied over onto the
scope-marker upstairs.
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
For (7), we propose an analysis that runs along
parallel lines the wie pronounced in the lower
clause has its scope assigned by a scope-marker
in the higher clause but whereas in (9) this
scope-marker is the bare wh- word wat, in (7)
the scope-marker shows concord with the wh-word
in the lower clause.
19
(7) wat zei Hans wiePers/Num/Gen zn fiets
had gestolen?
wie
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
For (7), we propose an analysis that runs along
parallel lines the wie pronounced in the lower
clause has its scope assigned by a scope-marker
in the higher clause but whereas in (9) this
scope-marker is the bare wh- word wat, in (7)
the scope-marker shows concord with the wh-word
in the lower clause.
20
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
To (8), the concordial scope-marking analysis
can be applied as well the only difference with
(7) is that this time there is no wh-element in
the embedded clause but a relative pronoun
instead.
This relative pronoun arguably occupies the same
slot in the structure as does the wh-pronoun in
(7).
21
(8) wat zei Hans diePers/Num/Gen zn fiets
had gestolen?
wie
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
To (8), the concordial scope-marking analysis
can be applied as well the only difference with
(7) is that this time there is no wh-element in
the embedded clause but a relative pronoun
instead.
This relative pronoun arguably occupies the same
slot in the structure as does the wh-pronoun in
(7).
22
(8) wat zei Hans diePers/Num/Gen zn fiets
had gestolen?
wie
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
Since the scope-marker originating in the higher
clause is itself inherently wh, the concord
relationship be- tween die and the scope-marker
is just what is needed to turn the scope marker
into wie.
The use of die in the lower clause in (8) is
really more economical than that of wie in the
lower clause in (7).
23
(8) wat zei Hans diePers/Num/Gen zn fiets
had gestolen?
wie
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(8') die zei Hans wie zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
The ungrammaticality of (8') follows from the
fact that the relative pronoun in the higher
clause fails to mark that clause as a question a
Dutch constituent question always needs a
wh-element at its left edge so (8')
is uninterpretable as a question, and there is no
other sensible interpretation available for it
either.
24
(10) wat zei Hans diePers/Num/Gen zn fiets
had gestolen?
X
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
Now that we have the outlines of an account for
(8), the analysis of (10) is straightforward it
is basically analogous to (8) the only
difference between them is that in (8) there is
concord between the die and the scope-marker
upstairs, whereas in (10) there is no such
concord.
25
(10) wat zei Hans diePers/Num/Gen zn fiets
had gestolen?
X
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(8') die zei Hans wie zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(9') wie zei Hans wat zn fiets had gestolen?
(10') die zei Hans wat zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
And of course (10') has the cumulative sins of
(8') and (9') combined the higher clause is
uninterpretable as a question, and the
scope-marker in the lower clause is entirely
redundant.
26
In sum, in none of the multiple exposure cases
is there movement of a wh-phrase from the lower
clause into the higher clause.
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
(10) wat zei Hans die zn fiets had gestolen?
Both scope-marking and wh-copying
constructions involve two separate
wh-dependencies one in the higher clause,
involving a scope-marker (which may or may not be
concordial), and another in the lower
clause, involving either a wh-element or (more
economically) a relative pronoun.
27
With this analysis of the multiple exposure
cases in place, lets now return to the example
in (6')
(6') wie zei Hans dat zn fiets had gestolen?
From (7)(10), we have learnt that Dutch can form
its long wh-dependencies via a variety of
double-Dutch strategies, none of which involve
long wh-movement.
This makes it logically possible to envision a
grammar for Dutch long wh-dependencies that does
not employ long wh-movement at all that is,
even (6') does not involve movement of wie from
the lower clause into the higher clause.
If (6') does not involve movement of wie from the
lower clause, the absence of a blocking effect
of dat follows.
28
If indeed (6') does not involve long wh-movement
of wie, it could instead be very much like (7) or
(8), with wie being a concordial scope-marker.
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(8) wie zei Hans die zn fiets had gestolen?
But while in (7) and (8) the lower clause
contains an overt element expressing at least the
person, number, and gender properties of the
wh-phrase, there is no such element in the lower
clause in (6'), which is intro- duced by the
subordinator dat.
29
So if (6') is to involve concordial
scope-marking rather than long wh-movement, what
is it that wie is in a concord relationship with
in this kind of sentence?
(6') wie zei Hans dat zn fiets had gestolen?
Our answer with itself that is, there is in
fact an instance of wie at the left edge of the
lower clause, but this wie is ultimately not
spelled out.
30
(7) wat zei Hans wiePers/Num/Gen zn fiets
had gestolen?
wie
Recall that in (7) and (8) the element
introducing the lower clause entertains a concord
relationship with the upstairs scope-marker for
its person, number, and gender properties.
Those properties are a subset of the full array
of features borne by these elements and as a
result, the concordial scope-marker and the
introducer of the lower clause are often not
strictly identical.
But imagine what would happen if there were not
a partial concord but a full concord relationship
between the scope-marker and the lower wh-element.
31
(11) wat zei Hans wieALL FEATURES zn fiets
had gestolen?
wie
What we end up with as a result of full concord
be- tween the two wh-elements is a situation in
which we have two fully identical wh-elements in
the syntactic structure, one looking down on the
other.
When we now want to linearise and spell out the
result of full concord in (11), we face a
dilemma one cannot precede or follow oneself.
To avoid a contradiction, the phonology realises
the output of the derivation in (11) by spelling
out only the higher of the two wh-elements.
32
(11) wat zei Hans wieALL FEATURES zn fiets
had gestolen?
wie
dat
With the lower wh-element left unrealised, we
would now seem to derive (6), which is
ungrammatical.
(6) wie zei Hans zn fiets had gestolen?
But (6) is independently rejected by the
requirement that the left edge of a subordinate
finite clause in Dutch may not be left empty.
An overt subordinating conjunction is therefore
needed whenever full-concordial scope-marking
takes place in a long wh-question in Dutch. So
(11) comes out as (6').
(6') wie zei Hans dat zn fiets had gestolen?
33
At this point, lets recall that in English
long-distance subject wh-questions, the
subordinating conjunction that must be absent in
the English versions of Dutch (6)/(6'), we find
that (5) is fine while (5') is bad.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
Previously, we had taken this to suggest that the
long- distance wh-question in (5) is derived by
movement of who from the lower clause into the
higher clause such movement cannot proceed
across that if the wh-phrase is a subject that
blocks the movement of who.
(5) who did Hans say (that) who stole his bike?
34
We will continue to subscribe to a
movement-based analysis of English long subject
questions the so- called that-trace effect
instantiates a ban on move- ment of a subject
wh-phrase across a conjunction.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
(5) who did Hans say (that) who stole his bike?
35
We will continue to subscribe to a
movement-based analysis of English long subject
questions the so- called that-trace effect
instantiates a ban on move- ment of a subject
wh-phrase across a conjunction.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
Technically, the that-trace effect can be
understood as follows.
The moved wh-phrase who leaves a silent copy
or trace of itself behind when it leaves the
lower clause.
(5) who did Hans say (that) who stole his bike?
36
We will continue to subscribe to a
movement-based analysis of English long subject
questions the so- called that-trace effect
instantiates a ban on move- ment of a subject
wh-phrase across a conjunction.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
This copy must find an antecedent within the
same clause but it doesnt find one when that
is present.
Movement out of the lower clause via a stopover
on the edge of that clause is ungrammatical
(5) who did Hans say (that) who stole his bike?
37
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
This copy must find an antecedent within the
same clause but it doesnt find one when that
is present.
Movement out of the lower clause via a stopover
on the edge of that clause is ungrammatical
(5) who did Hans say who that who stole his
bike?
38
NB this is often taken to be the standard way to
form long wh-movement dependencies but in my
recent work I have argued (on the basis of facts
from i.a. Hungarian and Chamorro) that it is in
fact ill-formed.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
This copy must find an antecedent within the
same clause but it doesnt find one when that
is present.
Movement out of the lower clause via a stopover
on the edge of that clause is ungrammatical
(5) who did Hans say who that who stole his
bike?
39
NB this is often taken to be the standard way to
form long wh-movement dependencies but in my
recent work I have argued (on the basis of facts
from i.a. Hungarian and Chamorro) that it is in
fact ill-formed.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
Movement to the edge of a clause is always
terminal onward movement from a clause edge is
impossible.
So long wh-movement must proceed in one fell
swoop, leaving an unlicensed copy of who when
that is there.
(5) who did Hans say who that who stole his
bike?
40
NB this is often taken to be the standard way to
form long wh-movement dependencies but in my
recent work I have argued (on the basis of facts
from i.a. Hungarian and Chamorro) that it is in
fact ill-formed.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
Movement to the edge of a clause is always
terminal onward movement from a clause edge is
impossible.
So long wh-movement must proceed in one fell
swoop, leaving an unlicensed copy of who when
that is there.
(5) who did Hans say (that) who stole his bike?
41
When that is not there, the structure of the
lower clause is reduced, and the copy does find a
local antecedent.
(5) who did Hans say stole his bike?
(5') who did Hans say that stole his bike?
Movement to the edge of a clause is always
terminal onward movement from a clause edge is
impossible.
So long wh-movement must proceed in one fell
swoop, leaving an unlicensed copy of who when
that is there.
(5) who did Hans say (that) who stole his bike?
42
English must form its long wh-questions via
movement from the lower clause into the higher
clause it does not have scope-marking or
wh-copying in its adult grammar.
Since adult English can only form its long
wh-questions via movement, it cannot avoid the
that-trace effect.
But children learning English as a first language
are well known to produce that-trace violations
as in (5').
(5') who did Hans say that stole his bike?
bad for adults fine for young children
(12) what do you think who is gonna jump?
(13) who do you think who is gonna jump?
Interestingly, studies on childrens acquisition
of wh- dependencies in English (Thornton 1990,
McDaniel, Chiu Maxfield 1995) report that
children who produce and accept (5') also produce
and accept scope-marking and/ or wh-copying
constructions of the type in (12)(13).
43
McDaniel et al. elicited grammaticalitity
judgements for (5'), (12) and (13) from children
aged 21157, using the following protocol
If someone kissed Grover, but we dont know
who, does it sound OK if I asked Nelly this way
Nelly, who do you think that kissed Grover?
(5') who did Hans say that stole his bike?
bad for adults fine for young children
(12) what do you think who is gonna jump?
(13) who do you think who is gonna jump?
44
McDaniel et al. elicited grammaticalitity
judgements for (5'), (12) and (13) from children
aged 21157, using the following protocol
If someone kissed Grover, but we dont know
who, does it sound OK if I asked Nelly this way
Nelly, what do you think who kissed Grover?
(5') who did Hans say that stole his bike?
bad for adults fine for young children
(12) what do you think who is gonna jump?
(13) who do you think who is gonna jump?
45
McDaniel et al. elicited grammaticalitity
judgements for (5'), (12) and (13) from children
aged 21157, using the following protocol
If someone kissed Grover, but we dont know
who, does it sound OK if I asked Nelly this way
Nelly, who do you think who kissed Grover?
(5') who did Hans say that stole his bike?
bad for adults fine for young children
(12) what do you think who is gonna jump?
(13) who do you think who is gonna jump?
They found that in all cases in which a child
accepted (12) or (13), the child also accepted
(5') on the other hand, children lacking
(12)/(13) tended to reject (5').
accept (5') reject (5') children with
(12)/(13) 37 0 children w/o
(12)/(13) 24 42
46
This correlation between allowing that-trace
sequences and having scope-marking and/or
wh-copying follows directly from what we have
talked about.
English-speaking children who allow that-trace
sequences have the grammar of adult Dutch for
some time (80 of youngest kids, 56 of middle,
17 of oldest).
(5') who did Hans say that stole his bike?
bad for adults fine for young children
(12) what do you think who is gonna jump?
(13) who do you think who is gonna jump?
(6') wie zei Hans dat zn fiets had gestolen?
(7) wie zei Hans wie zn fiets had gestolen?
(9) wat zei Hans wie zn fiets had gestolen?
47
Conclusion
What we have found, then, is that there are at
least two strategies to form long wh-dependencies
One strategy is to represent them structurally in
terms of three local dependencies one in the
higher clause, involving local movement of a
scope-marker, one in the lower clause, involving
local movement of the wh-phrase associated to the
scope-marker, and one linking the scope-marker to
its associate, facilitating concord.
The other strategy involves long-distance
movement of the wh-phrase itself, from the lower
clause into the higher clause.
48
Conclusion
Both strategies to form long wh-dependencies
have their own pros and cons
The concordial scope-marking strategy has the
benefit of representing an apparent long-distance
dependency in the form of a series of local
dependencies local dependencies are less costly
(read easier for the parser to deal with) than
long-distance dependencies. But this benefit is
offset by the fact that this strategy needs to
exploit two movements plus concord.
The long-distance movement strategy has the
benefit of exploiting just a single movement
operation. But this benefit is offset by the fact
that this movement operation creates a costly
long-distance dependency.
49
Conclusion
Long-distance wh-dependencies always come at a
cost. Different grammars put the cost in
different accounts.
Dutch and child-English prefer to avoid the
formation of literal long-distance movement
dependencies, by exploiting concordial scope
marking. Concomitantly, since Dutch and
child-English do not literally move a wh-phrase
out of a lower clause into a higher one, they
have no trouble with that-trace sequences.
Adult English, on the other hand, prefers not to
use concordial scope-marking dependencies, and
must instead perform long-distance wh-movement.
Concom- itantly, since long wh-movement of a
subject across a lexical conjunction is
prohibited, adult English shows that-trace
effects.
50
Thank you very much.

• Marcel den Dikken
• Mden-Dikken_at_gc.cuny.edu
• Hans Bennis
• hans.bennis_at_meertens.knaw.nl