Extraction out of subjects in German

Download Report

Transcript Extraction out of subjects in German

The Importance of Being a Complement:
CED Effects Revisited
Johannes Jurka
Dissertation Defense
June 3, 2010
What this dissertation is about
 Revisiting subject island effects cross-linguistically – in
particular alleged counterexamples
 Acceptability judgment studies in German, English, Japanese,
Serbian
 Various factors contribute to the acceptability of
extraction but complement/non-complement
asymmetry is real (CED)
Subject islands
Ross (1967): Sentential subject constraint
(1) *The teacher who [S that the principal would fire __ ]
was expected by the reports is a crusty old battleaxe.
Chomsky (1973): Subject condition
Subjects are islands in general
(2) *Who did [NP stories about__] terrify John?
CED and Freezing Effects
Condition on Extraction Domains: (Huang 1982:505)
a phrase A may be extracted out of a domain B only if
B is properly governed.
 all non-complements are islands (subjects, adjuncts)
Freezing Effects (Wexler & Culicover 1981)
no extraction is possible out of previously moved
domains
CED
*1982 – ✞2007 ?
 the CED is empirically inaccurate: extraction out of in-situ
subjects is grammatical  SI effects can be reduced to Freezing
Effects (Stepanov 2007)
 highly problematic for theories that try to derive a
complement/non-complement asymmetry:
 Barriers (Chomsky 1986)
 Multiple Spell-Out (Uriagereka 1999, Nunes & Uriagereka 2000)
But the argument is based on controversial data
Outline of this talk
 Methodology
 Experiments in German
 Experiment 1: NP-subextraction out of in-situ vs. derived subjects
 Experiment 2: NP-subextraction internal vs. external subjects
 Summary of other experiments (indirect objects, wh-extraction)
 Grammaticality vs. Acceptability
 Summary of other experiments in English, Japanese, Serbian
 ATB –was für split – ameliorates islands  derivation in
terms of sidewards movement
 Conclusion: CED holds and independent of freezing
Methodology
 7-point scale
 Why not Magnitude Estimation?
 Likert scale data produces same amount of information & is less
noisy (Weskott & Fanselow 2008, Murphy & Vogel 2008, Goodall et al. 2010)
 Speakers impose categoricity on continuos scale – not immune
to floor/ceiling effects (Sprouse 2007)
 Choice of reference sentence not only affects absolute but also
relative ratings: X > Y with reference A but Y > X with
reference B (Sprouse 2007)
 Easier on both participants and experimenter
Methodology
 7 perfectly acceptable – 1 horribly unacceptable
 multiple informants, multiple lexicalizations, fillers, latin-
square design, etc.
 Linguistically naïve native speakers
 Studies conducted online (using Alex Drummond’s spellout
software)
Why German?
 German allows for subjects to stay in-situ (Grewendorf 1989, Haider 1993, Diesing
1992, Müller 2004)
Subject positions in German (den Besten 1985, Grewendorf
1989, Haider 1993, Diesing 1992, Müller 2004, to appear)
Why German?
 German allows for subjects to stay in-situ (Grewendorf 1989, Haider 1993, Diesing
1992, Müller 2004)
 Lack of agreement about licit extraction domains:
SpecTP
IO
SpecvP
Internal argument
den Besten 1985
N/A
✖
✖
✔
Diesing 1992
✖
N/A
✔
✔
Haider 1993
N/A
✔
✔
✔
Lutz 2001
N/A
✔
✔
✔
Müller 2010
✖
✖
✖
✔
Was für split
[Was für einen Mann] hast du gesehen?
what for a man have you seen
Was hast du [__ für einen Mann] gesehen?
‘What kind of man did you see?’
Standard construction used to thest NP-subextraction in
German (den Besten 1985, Diesing 1992, Haider 1993, Müller 2010)
Experiment 1
Was für split out of transitives
6 conditions:
 Split from Subjects vs. Objects
 Split from InSitu vs. Derived Domain
 2 non-split control conditions
paper questionaire, n=31
Control Conditions – No split
Subject
[Was für eine Ameise] hat denn den Beamten gebissen.
what for a ant
has indeed the clerk bitten
‘What kind of ant bit the clerk?’
Object
[Was für einen Beamten] hat denn die Ameise gebissen.
what for a
ant
has indeed the clerk bitten
‘What kind of clerk did the ant bite’
Subject, -moved
[Was] hat denn [t für eine Ameise] den Beamten gebissen.
What has indeed for an ant
the clerk
bitten
Subject, +moved
[Was] hat [t für eine Ameise] denn den Beamten gebissen.
What has for an ant
indeed the clerk
bitten
Object, -moved
[Was] hat denn die Ameise [t für einen Beamten] gebissen.
what has indeed the ant
for a
clerk
bitten
Object, +moved
[Was] hat denn [t für einen Beamten]die Ameise gebissen.
what has indeed for a clerk
the ant
bitten
Experiment 1 - Predictions
‘Freezing only’ predictions :
1 difference between moved and unmoved subject
2 no subject/object asymmetry
That’s not
what we find!
O, +spl, -mvd
S, +spl, -mvd
S, +spl, +mvd
 fake data
Experiment 1 results (n=31)
7
*
Sub/Obj: p<.001
Moved/InSitu: p<.001
Sub/Obj x Extraction: p<.001
6
5
4
*
3
*
2
1
S, -spl
O, -spl
O, +spl, -mvd S, +spl, -mvd O, +spl, +mvd S, +spl, +mvd
Experiment 1 results
2 constraints active in German
A.Extraction out of subjects is degraded (CED)
B. Extraction out of moved domains is degraded (freezing)
 A is not reducible to B
Constraints are cumulative:
split from unmoved objects (no violation) >
split from unmoved subjects (violating A) >
split from moved objects (violating B) >
split from moved subjects (violating A and B)
Acceptability vs. Grammaticality
 We can only measure acceptability judgments
 We are giving up on the assumption that the violation of a
grammatical constraint always leads to absolute
unacceptability (Keller 2000, Featherston 2005)
 The violation of a grammatical constraint leads to decrease in
acceptability up to floor effects
 Grammatical violation can be cumulative (e.g. CED +
freezing)
Experiment 2
Was für split out of external vs. internal arguments
4x2 design (8 conditions):
 Argument Type:
 transitive subjects
 transitive objects
 unaccusative subjects
 passive subjects
 +/- Extraction
Online study, n=37
7
6
5
unacc
pass
4
obj
unerg
3
2
1
-split
+split
7.00
7.00
F = 11.1
11.1
pF<=.001
p < .001
6.00
5.00
4.00
Unergative
F = 6.7
p = .0011
6.00
5.00
Unaccusative
4.00
Unergative
Object
3.00
3.00
2.00
2.00
1.00
-split
+split
1.00
-split
+split
7.00
7.00
F=8
p = .006
6.00
5.00
6.00
5.00
Unergative
4.00
unacc
4.00
Passive
3.00
3.00
2.00
2.00
1.00
-split
+split
pass
UNACC X PASS: F=0,p = 1
obj
UNACC X OBJ: F=.2,p = .677
PASS X OBJ: F=.23,p = .633
1.00
-split
+split
Further experiments in German
 Extraction out of direct objects > indirect objects
 Expected under CED if IOs are specifiers/adjunct but potentially
problematic for Larson (1988), Baker (1988) view as IOs as
complements of V
 Wh-Extraction out of sentential objects >
sentential subjects
Experiments in English
Are there cases of NP-subextraction out of subjects in English? Cases cited
in the literature almost all involve pied-piping (Ross 1967, Chomsky 2008, Levine
& Sag 2002)
NP-subextraction in English:
 subject/object asymmetry largely goes away \w piep-piping
 strong asymmetry \w p-stranding
(1a) About which topic did a book sway the voters? ≈
(1b) About which topic did John read a book?
(2a) Which topic did a book about sway the voters? <
(2b) Which topic did John read a book about?
Possible explanation: pied-piped PPs are not cases of genuine extraction
but hanging topic (cf. Broekhuis 2005)
Experiments in English
English – Active vs. Passive:
No difference between extraction out of active vs. passive
subjects – both strongly degraded
 in English (unlike German) subjects always raise to SpecTP
(position of lower copy immaterial for extraction)
Which politican did a book about cause a scandal last year? ≈
Which politican was a book about published last year?
Experiments in Japanese & Serbian
Long distance scrambling in Japanese (\w Chizuru Nakao & Akira
Omaki)
 Subject sub-extraction = Object sub-extraction (cf. Lasnik&Saito 1992)
 Subject baseline vs. Object baseline
 Subject > Object (due to extra-grammatical factors)
 The degree of degradation is BIGGER in sub-extraction from Subject
than from Object  subject island effect exists in Japanese
NP-subextraction in Serbian (\w Ivana Mitrović):
 PP-extraction: Object > Subject
 Left Branch Extraction: Object = Subject
Interim Summary
 Crosslinguistic studies indicate that complements are the
preferred extraction domains
 CED violation does not necessarily yield full unacceptability
 CED & freezing are separate (cumulative) constraints
Subject island repair:
(1a) *?Was hat denn [SUB t für eine Prinzessin] einen Frosch geküsst?
what has indeed t
for a princess a
frog kissed
‘What kind of princess kissed a frog?’
✖
Subject island repair:
(1b) Was hat denn [SUB t für eine Prinzessin] [DO t für einen Frosch] geküsst?
what has indeed
for a princess
for a frog kissed
‘What kind of princess kissed what kind of frog?’
Licit gap in DO licenses illicit gap in SUB
✔
Indirect object island repair
(2a) *Was hast du denn [IO t für einem Professor] einen Studenten vorgestellt?
✖
what have you indeed
for a professor a student
introduced
‘What kind of professor did you introduce a student to?’
Indirect object island repair
(2b) Was hast du denn [IO t für einem Professor] [DO t für ein Student] vorgestellt?
✔
what have you indeed
for a professor
for a student introduced
‘What kind of professor did you introduce to what kind of student?’
Licit gap in DO liceneses illicit gap in IO
Subject island repair:
(1a) *?Was hat denn [SUB t für eine Prinzessin] einen Frosch geküsst?
what has indeed t
for a princess a
frog kissed
‘What kind of princess kissed a frog?’
✖
(1b) Was hat denn [SUB t für eine Prinzessin] [DO✖
t für einen Frosch] geküsst?
what has indeed
for a princess
for a frog kissed
‘What kind of professor was a student introduced to?’
Indirect Object island repair
✔
(2a) *Was hast du denn [IO t für einem Professor] einen Studenten vorgestellt? ✖
what have you indeed
for a professor a student
introduced
‘What kind of professor did you introduce a student to?’
(2b) Was hast du denn [IO t für einem Professor] [DO t für ein Student]
vorgestellt?✔
what have you indeed
for a professor
for a student introduced
‘What kind of professor did you introduce to what kind of student?’
ATB Was für Split
 Multiple question interpretation– one operator binding two
different variables: Opi… [xi … [yi … ]]
 ameliorates islands
 Amenable to PG/ATB analysis in terms of sideward
movement (Nunes&Uriagereka 2000, Hornstein&Nunes 2002)
 Derviation by Sidewards movement
What has for a princess for a frog kissed?
N={T, C, what, has, for, a, princess, for, a, frog, kissed}
What has for a princess for a frog kissed?
N={T, C, what, has, for, a, princess, for, a, frog, kissed}
side workspace
sideward movement
main workspace
SO of side workspace
 merge with main
spine
•Merging of all other elements
from the derivation (simplified)
• +Q-feature on C needs to be
checked
•Only lower copy of what accesible to
check Q-feature
•SO of highest copy by LCA (c-commands
all other copies)
movement
Voilà: Was hat für eine Prinzessin für einen Frosch geküsst?
(What has for a princess for a frog kissed?)
Presence of
lower copy
is crucial
Conclusion
 More careful look at the data: CED is emprically
accurate for German (and English, Japanese and possibly
universal) – violation of CED does not necessarily lead to
categorical unacceptability
 Additional freezing effects found – CED and freezing
cumulative in German
 Theoretical accounts for both CED and freezing need to be
maintained
Thank you!
Appendix
Experiment
 Wh-extraction out of non-finite sentential arguments in
German
Four conditions (2 x 2)
 Origin of Extraction: Subject vs. Object
 Movement: no wh-extraction vs. wh-extraction
Subject, no extraction:
[Die Diplomarbeit zu schreiben] hat die Studentin gelangweilt.
the
MA
to write
has the student
bored
‘Writing the MA bored the student.’
Subject, extraction:
Welche Arbeit hat denn [t zu schreiben]die Studentin gelangweilt?
which paper
has indeed to write
the student bored
‘Which paper did writing bore the student?’
Object, no extraction:
Die Studentin hat [die Diplomarbeit zu schreiben] vorgehabt.
the student
has the MA
to write
planned
‘The student planned to write the MA.’
Object, extraction:
Welche Arbeit hat denn die Studentin [t zu schreiben] vorgehabt?
which paper
has indeed the student
to write
planned
‘Which paper did the student plan to write?’
Non-finite clauses (n=31)
7.00
*
SUB/OBJ x EXTRACTION: p <.001
*
6.00
5.00
4.00
Subject
Object
3.00
2.00
1.00
-extraction
+extraction
Unaccusatives vs. Unergatives
2x2 design
Factors:
PREDICATE
EXTRACTION
7-point scale
n=27
Unaccusatives vs. Unergatives
7
6
PRED X EXT
F = 74.714
p < .001
5
4
Unacc
Unerg
3
2
1
-ext
+ext
Unaccusatives vs. Unergatives –
Individual Distribution
n=37
Marginal/no difference
30%
Unerg > Unacc
8%
Unacc > Unerg
62%
23…pattern of
mean
3…opposite pattern
11…no/marginal
difference
(mean
for 3 people
with
difference
≤.66)
opposite pattern:
2.33 vs.1
3 vs. 2
2.67 vs. 1.67
Passivized
ditransitives
factors:
• ± MARKED
• SUB/IO
+ 2 non-extraction
controls
n=23
7
6
5
4
3
2
1
No Scr
Scr
Sub
IO
Sub, Scr
IO, Scr
Experiment 3
Methods
・ 7-point scale acceptability judgment task
・ Subjects: Adult Native Japanese speakers (n=27)
Four conditions (2 x 2)
 Origin of Extraction: Subject vs. Object
 Movement: No scrambling vs. scrambling
Japanese conditions
(i) –scrambling, subject
少女は [ 姉が
ぬいぐるみを
捨てた こと が
girl-Top
sister-Nom stuffed animal-Acc dumped fact -Nom
けんかの原因 だ と]
訴えた。
fight-gen-cause Cop Comp claimed
“The girl claimed that [the fact that her sister dumped her stuffed animal] is
the cause of the fight”
(ii) +scrambling, subject
ぬいぐるみを1
少女は
[ 姉が
t1 捨てた こと が
stuffed animal-Acc girl-Top
sister-Nom
dumped fact -Nom
けんかの原因 だ と]
訴えた。
fight-gen-cause Cop Comp claimed
“Her stuffed animal1, the girl claimed that [the fact that her sister dumped
t1]is the cause of the fight”
Japanese conditions
(iii) –scrambling, object
少女は [姉が2
PRO2 ぬいぐるみを
捨てた こと を
girl-Top sister-Nom
stuffed animal-Acc dumped fact -Acc
内緒にしていた と]
訴えた。
secret-Dat-kept Comp claimed
“The girl claimed [that her sister2 kept as a secret [the fact that she2 dumped
her stuffed animal]].”
(iv) +scrambling, object
ぬいぐるみを1
少女は [姉が2
PRO2 t1 捨てた こと を
stuffed animal-Acc girl-Top sister-Nom
dumped fact -Acc
内緒にしていた と]
訴えた。
secret-Dat-kept Comp claimed
“Her stuffed animal1, the girl claimed [that her sister2 kept as a secret [the fact
that she2 dumped t1]].”
Japanese data N = 27
7.00
*
SUB/OBJ X EXTRACTION: p <.001
6.00
5.00
4.00
ns
3.00
2.00
1.00
-extraction
+extraction
Subject
Object
Another look at the sentences
(i), (ii): Extraction out of an embedded subject complex NP
girl-Top [ sister-Nom stuffed animal-Acc dumped fact -Nom fight-gen-cause
Cop Comp] claimed
“The girl claimed that [the fact that her sister dumped her stuffed animal]is the cause of the
fight”
(iii), (iv): Extraction out of an embedded object complex NP
girl-Top [sister-Nom2
PRO2
stuffed animal-Acc dumped fact -Acc
secret-Dat-kept Comp] claimed
“The girl claimed [that her sister2 kept as a secret [the fact that she2 dumped her stuffed
animal]].”
Object conditions involve center-embedding and 3
subjects in a row: Extremely hard to process (e.g., Babyonyshev &
Summary
 Subject sub-extraction vs. Object sub-extraction
 Subject = Object
 Replicates pattern reported in Lasnik & Saito 1992
 Subject baseline vs. Object baseline
 Subject > Object (due to center embedding)
 The degree of degradation is BIGGER in sub-extraction from Subject
than from Object
 Subject Condition holds in Japanese
7
Fillers
No floor effect!
6
5
4
3
2
1
long-distance
scrambling
Subj, Move
Obj, Move
Center-embedding why' in-situ inside
RC
CSC violation
Melting effects (de Hoop 1996, Müller to appear)
Extraction out of external subjects is possible only if object
is scrambled across the subject
7.00
6.00
*
5.00
*
ns
4.00
3.00
2.00
1.00
sub
sub, scrambled obj
obj
Melting
 The ‘melting’ effect does not hold as reported – scrambling
of object does not improve acceptability of extraction out of
external subjects
 Again both subject extraction conditions worse than
extraction out of object
 Follow-up studies needed to further investigate the role of
information structure (scrambling of object results in marked
word) – testing acceptability in context
The LCA in SOV languages
 Kayne (1994): all languages are S-H-C – S-C-H structures
are derived through overt movement of the C
 That would predict that extraction out of objects should be
generally disallowed – contrary to the German data
 How can reconcile this while maintaining an MSO in S-H-C
structures?
 Linear ordering is lexically determined by the head.
The LCA only applies to determine the ordering
between two non-heads.
{H,C} order not determined by LCA
Technical argument: Head/complement
asymmetry stipulated by Kayne
Empirical argument: many languages
show mixed head/complement
patterns in different domains
(German, Dutch, Hindi) –
Or even with the same category with
different lexical items:
Was hat denn für ein Käfer den Beamten gebissen?
what has for a beetle the clerk bitten
Specifier assembled in separate workspace
Specifier is first
linearized and
then merged into
main spine 
not accesible for
further syntactic
operations, i.e.
island
main spine