ABSTRACT
Title of dissertation: THE IMPORTANCE OF BEING A COMPLEMENT
CED-EFFECTS REVISITED
Johannes Jurka, Doctor of Philosophy, 2010
Dissertation directed by: Professor Norbert Hornstein
Department of Linguistics
This dissertation revisits subject island effects (Ross 1967, Chomsky 1973) cross-
linguistically. Controlled acceptability judgment studies in German, English, Jap-
anese and Serbian show that extraction out of specifiers is consistently degraded
compared to extraction out of complements, indicating that the Condition on Ex-
traction domains (CED, Huang 1982) is still empirically viable, contrary to recent
claims (Stepanov 2007). As a consequence, recent treatments of the CED in terms
of Multiple Spell-Out (Uriagereka 1999) are still tenable. First, a series of NP-
subextraction experiments in German using was f?ur-split is discussed. The results
indicate that subject island effects cannot be reduced to freezing effects (Wexler &
Culicover 1981). Extraction out of in-situ subjects is degraded compared to extrac-
tion out of in-situ objects. Freezing incurs an additional cost, i.e., extraction out
of moved domains is degraded compared to in-situ domains. Further results from
German indicate that extraction out of in-situ unaccusative and passive subjects
is en par with extraction out of objects, while extraction out of in-situ transitive
and intransitive unergative subjects causes a decrease in acceptability. Additionally,
extraction out of indirect objects is degraded compared to extraction out of direct
objects. It is also observed that a second gap improves the acceptability of otherwise
illicit was f?ur-split, a phenomenon dubbed Across-the-Board (ATB)-was f?ur-split
and analysed in terms of Sideward Movement (Hornstein & Nunes 2002). Further-
more, wh-extraction out of non-finite sentential arguments also shows a significant
subject/object asymmetry. Experiments in English indicate that NP-subextraction
yields the familiar subject/object asymmetry, while the contrast largely disappears
when PPs are fronted. Further results show that ECM and passive predicates do not
improve the acceptability of the extraction out of subjects. Finally, subject subex-
traction patterns in Japanese and Serbian are investigated. Both Long-distance
scrambling and clefting out of sentential subjects in Japanese leads to a stronger
degradation than out of sentential objects. PP-extraction in Serbian also shows the
same subject/object asymmetry, while no such contrast is found for Left Branch
Extraction.
THE IMPORTANCE OF BEING A COMPLEMENT
CED-EFFECTS REVISITED
by
Johannes Jurka
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2010
Advisory Committee:
Professor Norbert Hornstein, Chair
Professor Juan Uriagereka
Professor Amy Weinberg
Assistant Professor Valentine Hacquard
Associate Professor Susan Dwyer
c? Copyright by
Johannes Jurka
2010
Dedication
Ich widme diese Arbeit meiner Oma Herta Andrysek.
ii
Acknowledgments
First, I would like to thank the Academy ... seriously, I would like to thank the
Austrian Academy of Science for giving me copious amounts of money to work on
this dissertation (DOC dissertation grant 22397).
I?d like to thank my committee: Norbert Hornstein, Juan Uriagereka, Amy
Weinberg, Valentine Hacquard and Susan Dwyer. Norbert for being a mentor, a
coach or simply the best advisor possible. Norbert gave me all the guidance I asked
for while leaving me all the freedom I needed. Without him, this dissertation would
not exist. Juan for getting me excited about this topic and introducing me to the
data that made me run all these experiments. Amy for stepping in to be on my
committee in the last moment. Valentine for extremely useful comments and Susan
Dwyer for agreeing to be the dean?s representative.
I?d also like to thank Howard Lasnik, who unfortunately could not be on the
committee but gave me very helpful comments anyway, for always having an open
door for his students (quite literally), for being one of the greatest teachers I have
ever had and a fierce ping-pong player.
So many people at the UMD linguistics department have made this the perfect
environment to write a dissertation. Special thanks go to Akira Omaki for teaching
me most of what I know about doing experiments. Alex Drummond for helping me
with the online studies. Maki Kishida for always patiently assisting me with my silly
LATEX questions. Kathi Faulkingham, Kim Kwok and Robert Magee for showing me
the way through the jungle of I-20s, I-94s, W2s, W4s, SEVIS confirmations, travel
requests, non-resident alien W-9s etc. ...
iii
So many colleagues and friends at the UMD linguistics department have made
the College Park life worth living. I would like to thank Tim Hawes for all the
lunches and dinners and the conversations about everything but linguistics. Alexan-
der Williams for his admiration of Marianne Mendt and for being arguably the
funniest story teller ever. Michael Gagnon and Brad Larson for making it close to
impossible to get work done in 1407. Alexis Wellwood, Annie Gagliardi, Ariane
Rhone, Brian Dillon, Chris Laterza, Dan Parker, Dave Kush, Diogo Almeida, Ellen
Lau, Eri Takahashi, Ewan Dunbar, Greg Cogan, Ilknur Oded, Josh Riley, Matt Wa-
gers, Megan Sutton, Pedro Alcocer, Phil Monahan, Shevaun Lewis, Shiti Malhotra,
Terje Lohndal, Tim Hunter, Yakov Kronrod - and this is by no means an exhaustive
list - have all contributed in their own very special ways to making this a truely
enjoyable time.
I?d especially like to thank all the students I had the honor of teaching.
Outside of linguistics, I?d like to thank Ricky Pugliese for being such a great
friend and always putting me up when I needed a place to stay (be it in DC, Venice,
CA. or Venice, Italy). Juliana Araujo and Malthe Borch for being there for me
pretty much since day one of my American experience. Seth Lawler for being an
awesome roommate and for introducing me to Jazz.
The following is a list of things that have also contributed immensely to my
physical and mental well-being while writing this dissertation: Curb your Enthusi-
asm, HR57, Kottan, Martina Sagmeister?s Impro Dance class, Die Supernacht der
Weihnachtsstars,Tranquil Space Yoga, the UMD Squash Club, The Wire.
Ich rate es niemanden, ?uber eine l?angeren Zeitraum auf zwei Kontinenten gle-
iv
ichzeitig zu leben. Ich bin trotzdem froh, dass ich?s gemacht hab und alle lieben
Menschen in Wien auch oft sehen konnte. Ich bedanke mich bei Christoph Thal-
hamer fuer die zahlreichen nihilistisch-linguistisch-philosophischen, philosophisch-
nihilistisch-linguistischen sowie linguistisch-philosophisch-nihilistischen Gespr?ache.
Christian Url, Claudia Schuster, Georg Blume, Marton Matura, Peter Krauss,
Robert Adelmayer, Severin Dostal und Veronika Selisko f?ur die jahrelange Freund-
schaft.
Vielen Dank an meine Eltern, die mich immer bedingslos unterst?utzt haben.
My deepest love goes to Amelia Wein. I?m so glad we?ve met! Thanks for your
love, patience, inspiration and support in every way possible. I wouldn?t be who I
am without you.
v
Contents
List of Tables x
List of Figures xi
1 Introduction 1
2 Background 9
2.1 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Descriptive Generalizations . . . . . . . . . . . . . . . . . . . 9
2.1.1.1 The (?*Sentential) Subject Constraint . . . . . . . . 9
2.1.1.2 The Condition on Extraction Domains (CED) . . . . 11
2.1.1.3 Freezing Effects . . . . . . . . . . . . . . . . . . . . . 12
2.1.2 Theoretical proposals . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2.1 Multiple Spell-Out (MSO) . . . . . . . . . . . . . . . 14
2.1.2.2 Subject island effects as Freezing Effects . . . . . . . 20
2.2 The inadequacy of informal judgments . . . . . . . . . . . . . . . . . 21
2.2.1 De iudiciis non est disputandum? . . . . . . . . . . . . . . . . 22
2.2.2 Operational procedures . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Arrogance and modesty . . . . . . . . . . . . . . . . . . . . . 27
2.2.4 Judgment = Grammar + X . . . . . . . . . . . . . . . . . . . 29
2.2.4.1 Capturing gradience . . . . . . . . . . . . . . . . . . 32
2.2.5 Christmas trees . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 NP-Subextraction in German 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.1 What?s a subject? . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.2 Subject positions in German . . . . . . . . . . . . . . . . . . . 39
3.1.2.1 SpecCP . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.2.2 SpecvP and SpecTP . . . . . . . . . . . . . . . . . . 40
3.1.2.3 Internal and external subjects . . . . . . . . . . . . . 46
3.1.3 A brief history of German NP-subextraction . . . . . . . . . . 47
3.1.3.1 den Besten 1985 . . . . . . . . . . . . . . . . . . . . 49
3.1.3.2 Melting . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.3.3 Diesing 1992 . . . . . . . . . . . . . . . . . . . . . . 52
3.1.3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Experiment 1 - External vs. derived subjects . . . . . . . . . . . . . . 56
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.2 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.3.1 Participants . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.3.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.3.3 Design . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
vi
3.2.4.1 Distribution of the data . . . . . . . . . . . . . . . . 61
3.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.5.1 Excluding extra-grammatical factors . . . . . . . . . 66
3.2.5.2 Why are things as bad as they are but not worse? . . 70
3.3 Experiment 2 - Passivized Ditransitives . . . . . . . . . . . . . . . . . 72
3.3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.2 Methodology and Design . . . . . . . . . . . . . . . . . . . . . 73
3.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.4.1 How to interpret Likert-scale results . . . . . . . . . 75
3.3.4.2 Analysis of the effects . . . . . . . . . . . . . . . . . 77
3.3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4 Experiment 3 - Internal and external arguments . . . . . . . . . . . . 86
3.4.1 Experiment 3A . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 86
3.4.1.2 Methodology and Design . . . . . . . . . . . . . . . . 86
3.4.1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4.1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 91
3.4.2 Experiment 3B . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.4.2.1 Methodology and Design . . . . . . . . . . . . . . . . 93
3.4.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4.2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.5 Experiment 4 - Melting . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.5.2 Methodology and Design . . . . . . . . . . . . . . . . . . . . . 99
3.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.6 Accounting for the data . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.6.1 The LCA in SOV languages . . . . . . . . . . . . . . . . . . . 104
3.6.1.1 The LCA as last resort . . . . . . . . . . . . . . . . . 107
3.6.2 A MSO Account of was-f?ur split . . . . . . . . . . . . . . . . 110
3.7 ATB was-f?ur split . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.7.1 A sideward movement account of ATB was-f?ur split . . . . . 117
3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4 Extraction out of non-finite clauses in German 122
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.3 Experiment 5 - Sentential subjects vs. sentential objects . . . . . . . 127
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.3.2 Methodology and design . . . . . . . . . . . . . . . . . . . . . 127
4.3.3 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
vii
4.3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.4 Experiment 6 - Auxiliaries vs. Main verbs . . . . . . . . . . . . . . . 135
4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.4.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 137
4.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.5 Experiment 7 - Separable verbs . . . . . . . . . . . . . . . . . . . . . 141
4.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.5.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 142
4.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5 Subextraction in English 147
5.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.2 Experiment 8 - Pied-piping in English . . . . . . . . . . . . . . . . . . 151
5.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.2.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 153
5.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2.3.1 Distribution of the data . . . . . . . . . . . . . . . . 156
5.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3 Experiment 9 - ECM . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 160
5.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.4 Experiment 10 - Passives . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.4.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 165
5.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.5 Experiment 11 - Subject islands and reconstruction . . . . . . . . . . 168
5.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.5.2 Design and Methodology . . . . . . . . . . . . . . . . . . . . . 170
5.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.5.3.1 Distribution of the data . . . . . . . . . . . . . . . . 173
5.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6 A glance at other languages: Subextraction in Japanese and Serbian 177
6.1 Japanese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.1.2 Experiment 12 - Scrambling and Clefting . . . . . . . . . . . . 181
viii
6.1.2.1 Design and Methodology . . . . . . . . . . . . . . . . 181
6.1.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . 183
6.1.2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 184
6.1.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.2 Serbian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.2.2 Experiment 13: Left Branch Extraction vs. PP-extraction . . 189
6.2.2.1 Design and Methodology . . . . . . . . . . . . . . . . 189
6.2.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.2.2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 192
7 Conclusion 193
A Instructions 196
A.1 German . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
A.2 English . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
A.3 Serbian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
A.4 Japanese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
B Items 205
B.1 Experiment 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
B.2 Experiment 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
B.3 Experiment 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
B.3.1 Experiment 3A . . . . . . . . . . . . . . . . . . . . . . . . . . 207
B.3.2 Experiment 3B . . . . . . . . . . . . . . . . . . . . . . . . . . 209
B.4 Experiment 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
B.5 Experiment 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
B.6 Experiment 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
B.7 Experiment 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
B.8 Experiment 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
B.9 Experiment 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
B.10 Experiment 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
B.11 Experiment 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
B.12 Experiment 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
B.13 Experiment 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
ix
List of Tables
3.1 Which domains allow subextraction in German? . . . . . . . . . . . . 56
3.2 Experiment 1: Descriptive statistics . . . . . . . . . . . . . . . . . . . 63
3.3 Experiment 1: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 Experiment 2: Descriptive Statistics . . . . . . . . . . . . . . . . . . . 76
3.5 Experiment 2: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.6 Experiment 3: Descriptive Statistics . . . . . . . . . . . . . . . . . . . 88
3.7 Experiment 3: ANOVA 2x2 subanalyses . . . . . . . . . . . . . . . . 90
4.1 Experiment 5: Means . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Experiment 5: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.3 Experiment 5: ANOVA subanalysis for sentential subjects . . . . . . 133
4.4 Experiment 5: ANOVA subanalysis for sentential objects . . . . . . . 133
4.5 Experiment 6: Descriptive Statistics . . . . . . . . . . . . . . . . . . . 140
4.6 Experiment 6: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.7 Experiment 7: Descriptive Statistics . . . . . . . . . . . . . . . . . . . 144
4.8 Experiment 7: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.1 Experiment 8: Descriptive results . . . . . . . . . . . . . . . . . . . . 154
5.2 Experiment 8: 2x2 ANOVA subanalyses . . . . . . . . . . . . . . . . 155
5.3 Experiment 8: Means . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.4 Experiment 8: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.5 Experiment 10: Means . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.6 Experiment 10: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.7 Experiment 11: Means . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.8 Experiment 11: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.1 Experiment 12: Descriptive Statistics . . . . . . . . . . . . . . . . . . 183
6.2 Experiment 12: ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . 183
x
List of Figures
2.1 Christmas Trees in front of the White House and Rockefeller Center . 34
3.1 Experiment 1: Predictions . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2 Experiment 1: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 Experiment 1: Individual results of all 31 participants . . . . . . . . . 64
3.4 Experiment 2: Means . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.5 Subextraction out of IOs - individual distribution . . . . . . . . . . . 79
3.6 Experiment 3: Histograms . . . . . . . . . . . . . . . . . . . . . . . . 83
3.7 Experiment 3A: Results . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.8 Experiment 3B: Results . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.9 Experiment 4: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.10 The Olympic podium . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.1 Experiment 5: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Experiment 6: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.3 Experiment 7: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.1 Experiment 8: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.2 Experiment 8: Histograms . . . . . . . . . . . . . . . . . . . . . . . . 157
5.3 Experiment 8: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.4 Experiment 10: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.5 Experiment 11: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.6 Experiment 11: Histograms . . . . . . . . . . . . . . . . . . . . . . . 173
6.1 Experiment 12: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.2 Scrambling conditions compared with filler items . . . . . . . . . . . . 186
6.3 Experiment 13: Results . . . . . . . . . . . . . . . . . . . . . . . . . . 191
xi
Wir f?uhlen, dass, selbst wenn alle m?oglichen wissenschaftlichen Fragen
beantwortet sind, unsere Lebensprobleme noch gar nicht ber?uhrt sind. Freilich
bleibt dann eben keine Frage mehr; und eben dies ist die Antwort.
? Ludwig Wittgenstein, Tractatus Logico-Philosophicus, 6.52 ?
xii
Chapter 1
Introduction
Wh-extraction is one of the most intriguing and most widely studied phenomena in
theoretical linguistics. In a language like English, a wh-phrase surfaces in a different
position from where it is interpreted:
(1) a. Mary kissed John.
b. Whoi did Mary kiss ei?
The dependency between the surface position of the wh-element and the position of
interpretation, known as wh- or ?A-dependency, can span over a potentially infinite
distance:
(2) Whoi did Kim say that Bill thought that Tim believed that Ben kissed ei ?
There are certain configurations, however, that seem to interfere with the formation
of a wh-dependency:
(3) a. *Whoi did a book about ei cause a scandal?
b. *Whati did you eat ei and fries?
c. *Which book did you fall asleep after you read ei?
Configurations of this sort were dubbed islands by Ross (1967), who invokes the
image of the wh-phrase being stranded in a position from which it cannot escape.
This dissertation is about a specific subset of islands, namely subject islands of the
sort exemplified in (3-a). We will approach this issue primarily from an empirical
1
point of view addressing the question of how robustly the generalization holds that
subjects constitute islands cross-linguistically.
Chapter 2 provides the background for the discussions to ensue. We will fast-
forward through the history of subject islands. It all started, as it so often does,
with Ross?s (1967) dissertation. Ross formulated the Sentential Subject Constraint,
according to which extraction out of NPs immediately dominated by S is disallowed.
Chomsky (1973) extended these observations to a general Subject Condition, bar-
ring subextraction out of subjects across the board. Subsequently, Cattell (1976)
and more famously Huang (1982) subsumed subject island effects under a general
condition on extraction out of non-complements. This idea has become known as
Huang?s Condition on Extraction Domains (CED).
Over the years, unsparing efforts have been made to capture CED-effects. Two
accounts deserve special mention. Government and Binding Theory has given us
Barriers (Chomsky 1986), where CED-effects are incoporated into a general theory
of locality. For Chomsky, the complement/non-complement asymmetry is ultimately
a function of lexical marking or L-marking, which is roughly defined as being ?-
marked by a lexical head. Only complements are L-marked, and only complements
are licit domains for extraction. The core minimalist proposal is due to Uriagereka
(1999), who derives CED-effects from general constraints on the structure building
mechanism.
So far so good. Ever since Ross?s dissertation, however, apparent counterexam-
ples to the subject condition and consequently to the CED from various languages
have been floating around. Stepanov (2001, 2007) has collected some of these cases
and proposes that subjects are not islands in general, but that the ungrammatical-
ity of many examples is the result of freezing effects (cf. Wexler & Culicover 1981,
Takahashi 1994). In other words, nothing is wrong with extracting out of subjects
per se but subjects become opaque domains as a result of being moved. Languages
2
that allow for subjects to stay in-situ, so Stepanov?s argument continues, therefore
allow extraction out of subjects. Stepanov?s line of reasoning seems compelling, and
if he is right about the data, this would be rather dire news for the above mentioned
theories that try to derive subject island effects independent of freezing effects.
It seems to be the case in science generally that a theory is only as good as
the data it is built on, while, conversely, the data we discover is only as good as the
theoretical assumptions we hold when looking for it. The situation is no different
with syntactic islands. While much insightful theoretical work has been done on
islands and the CED, the status of the empirical facts the theories are based on is
often highly controversial. Key data points used by Stepanov to build his argument
are of this sort.
In this thesis, we revisit many of the apparent counterexamples to the CED
and employ more-fine grained methods of data collection. Concretely, we conducted
a series of 7-point scale acceptability judgment studies in German, English, Japanese
and Serbian, following a much more rigid methodological standard. These studies
quite uniformly converge on the conclusion that CED effects indeed exist but that a
violation of the CED does not always lead to categorical ungrammaticality.
The need for experimental data collection is emphasized in areas where key
data points are controversial. We will revisit such notions as grammaticality, ac-
ceptability and gradience in the grammar. Furthermore, we will discuss potential
sources of disagreement on judgments and how they can be reconciled. We will also
outline the methodology all of the studies discussed in this dissertation follow.
Chapter 3 focuses on NP-subextraction in German, employing the was-f?ur
split construction as a diagnostic to investigate what the extraction domains in
German are. The overarching question is whether German shows subject/object
asymmetries that are independent of freezing effects. We will first provide back-
ground on the notion of subject and subject position in German, concluding that
3
German has four positions where subjects can appear in overt syntax: a topic po-
sition in CP, a derived subject position in TP, an external subject position in Spec
vP and an internal subject position as the complement of V. We will then sum-
marize some of the theoretical work on subextraction in German, and will note a
considerable amount of disagreement on the status of key examples.
Experiment 1 looks at the acceptability of subextraction along two dimen-
sions: subjects vs. objects and in-situ vs. moved domains. The results indicate
that there is a significant subject/object asymmetry for in-situ domains. In other
words, there is a complement/non-complement asymmetry that is independent of
freezing. Freezing, in turn, is a separate factor that incurs an additional decrease
in acceptability. Extraction out of moved subjects and objects is worse than out
of their in-situ counterparts. We conclude that German has two independent con-
straints on extraction: the CED and freezing. This pattern cannot be captured by
a single theory but both a theory of the CED and of freezing effects is needed. In
addition, we will comment on the relation between a grammatical constraint and
the perceived acceptability of a sentence, addressing the issue of what it means for
a sentence to violate the CED and still be marginally acceptable.
Experiment 2 focuses on passivized ditransitives, which have the interesting
property that the subject linearly follows the indirect object in the unmarked word
order. We will see that was-f?ur split out of indirect objects is strongly degraded
compared to extraction out of internal subjects. This will lead us into an interesting
discussion on the structural properties of indirect objects in general. Addition-
ally, we find that manipulating the word order does not affect the permeability of
the domains. This suggests that an explanation of the acceptability patterns in
purely extra-grammatical terms, e.g. processing and information structure, cannot
be sufficient but that the difference in acceptability is a reflection of a grammatical
distinction.
4
Experiment 3 extends the domain of investigation to different predicate types.
We find that extraction out of internal arguments, i.e. subject of passives and unac-
cusatives and as well as objects, is preferred to extraction out of external arguments,
i.e. unergative subjects. A follow-up study confirms these results when intransitive
unaccusatives vs. unergatives are compared. Extraction out of unaccusative sub-
jects is preferred to extraction out of unergative subjects. Again, complements seems
to be the preferred domains of extraction.
Experiment 4 investigates a phenomenon M?uller (2010) refers to as Melting.
His claim is that extraction out of unergative subjects is licit when the object scram-
bles across the subject. We will see that the facts do not quite hold as presented by
M?uller . We concede, however, that a number of conflicting factors makes it difficult
to draw any stronger conclusions based on this study.
The bottom line of all our experiment reported in this chapter is that only
internal subjects, the complements of V, allow subextraction without any decrease
in acceptability. We will argue that these facts fall out from Uriagereka?s (1999)
Multiple Spell-Out theory. This will necessitate, however, a slight modification of
Kayne?s (1994) Linear Correspondence Axiom (LCA). Concretely, we will propose
that linearization is generally determined by a lexical feature on the head while the
LCA is only called upon in the elsewhere case.
We will conclude Chapter 3 by presenting a construction that, to my knowl-
edge, has not been discussed in the literature so far: Across-the-Board Was-f?ur
split, which has the property of ameliorating island violations in a way parallel to
Parasitic Gaps. We will propose that these facts can be captured by Nunes?s (2001)
and Nunes & Uriagereka?s (2000) treatment of Parasitic Gaps in terms of Sidewards
Movement.
Chapter 4 looks at extraction out of non-finite clauses in German. After
reviewing the literature on this topic, the results of our experiments again lead us
5
to the conclusion that extraction out of subjects is degraded compared to extraction
out of objects.
Experiment 5 directly compares non-finite sentential subjects and objects in
extraposed and non-extraposed position. On top of the subject/object asymmetry
that is diagnosed across the board, our results indicate that extraposition only has
a marginal effect on the permeability of a domain.
Experiment 6 and 7 pick up on an observation made by Grewendorf (1989),
according to which the validity of subject subextraction is determined by whether
the V2 position is filled with an auxiliary or a main verb. While Grewendorf?s
intuition is confirmed that there really is a main verb/auxiliary difference, this does
not only hold for subject subextraction but carries over to object subextraction
as well. The subject/object asymmetry persists, when this effect is controlled for.
Since object subextraction is affected by the main verb/auxiliary difference as well,
we speculate that this is a processing rather than a grammatical effect. Experiment
7 tries to shed light on the nature of this processing effect by investigating whether
separable particle verbs show the same pattern observed by Grewendorf.
Chapter 5 leaves German behind and looks at subject island effects in En-
glish. While many of the existing constraints were formulated largely on the basis
of English, a number of apparent counterexamples have been put forward in the
literature. This goes back to Ross?s formulation of the sentential subject condition,
which explicitly exempted NP-subextraction.
Experiment 8 takes Ross?s examples as its point of departure and indicates
that Pied-piping is major confounding factor with subextraction out of NPs. We
find a highly significant subject/object asymmetry when the preposition is stranded
but a much smaller difference when the preposition is pied-piped. We take this as
evidence for an analysis of fronted PPs as hanging topics that have not undergone
genuine subextraction. As a result, such constructions do not constitute counter-
6
evidence to the CED.
Experiment 9 then scrutinizes extraction out of subjects of ECM predicates,
which have sometimes been argued to be transparent for extraction (e.g. Chomsky
2008). Our results do not confirm these judgments but indicate that extraction
out of ECM complements is just as severely degraded as extraction out of regular
subjects. This comes as no surprise given that under plausible assumptions such
extractions both violate freezing and the CED.
Experiment 10 investigates the effect of different predicate types on the per-
meability of the subject, similarly to what experiments 2 and 3 did for German.
In English the situation is different in as far as the subject always undergoes rais-
ing to the SpecTP position. The question remains whether the grammar takes
the position of the lower copy into consideration when subextraction occurs, i.e. a
non-complement position for unergatives and a complement position for passives.
It turns out that this lower position is largely immaterial for the acceptability of
subextraction and that extraction out of any kind of subject is strongly degraded
in English.
Experiment 11, finally, scrutinizes an intriguing piece of data presented by
Sauerland & Elbourne (2002), who claim that subextraction out of passive subjects
is licit if the raising to SpecTP can be delayed until PF. This scenario occurs when
a second quantifier is present and the interpretation of the subject in its in-situ
position has a semantic effect. We do indeed find effects pointing in the direction
suggested by Sauerland & Elbourne, but we are forced to conclude that our results
neither fully support nor completely refute their claims.
Chapter 6 further extends the cross-linguistic coverage. Experiment 12, col-
laborative work with Chizuru Nakao and Akira Omaki, revisits extraction out of
non-finite sentential subjects in Japanese, which has often been claimed to be ac-
ceptable. We will point to a number of interfering factors that make it somewhat
7
cumbersome to test these cases in Japanese. Given, however, that we do find an in-
teraction between subjects/objects and extraction in favor of objects, we tentatively
conclude that Japanese does in fact show CED effects as well.
Experiment 13, based oncollaborativework withIvana Mitrovi?c, briefly touches
upon the difference between Left Branch Extraction (LBE) and PP-extraction(PPE)
in Serbian, a language where few controlled acceptability judgment studies have been
conducted so far. We find that while PPE shows the familiar CED effects, there
is no subject/object asymmetry for LBE. We take this to be evidence for analyses
of LBE in Slavic in terms of remnant movement (Ba?si?c 2004) or scattered deletion
(Cavar & Fanselow 2002).
Chapter 7 provides a conclusion and suggestions for future research. Ap-
pendix A lists the instructions to the experimental studies in German, English,
Serbian and Japanese. Appendix B provides the full set of stimuli of all 13 experi-
ments.
8
Chapter 2
Background
2.1 Theoretical Background
2.1.1 Descriptive Generalizations
2.1.1.1 The (?*Sentential) Subject Constraint
The original observation that there is something unruly about extracting out of
subjects goes back to Ross?s (1967) discussion of examples like (1):
(1) *The teacher who that the principal would fire was expected by the reports
is a crusty old battleaxe.
(Ross 1967, p. 241, ex. 4.251b)
Ross discards the generalization that ?reordering of subconstituents of subjects noun
phrases? (p.241) is illicit, arguing that such a rule would wrongly block examples
like (2-a) for which he provides the structure in (2-b) (p.242, ex. 4.253).
(2) a. Of which cars were the hoods damaged by the explosion?
9
b. S
Q NP
NP
the hoods
NP1
P
of
NP2
which cars
VP
were damaged by the explosion
In (2) the NP2 of which car is subextracted from inside the subject NP1. Ross
observes that (1) differs from (2) in that the subject NP in the former is also dom-
inated by an S, which he takes to be the crucial difference. He concludes that the
following constraint holds for English:
(3) The Sentential Subject Constraint (SSC)
No element dominated by an S may be moved out of that S if that node S is
dominated by an NP which itself is immediately dominated by S.
Even though Ross avoids the term subject in the definition of his constraint, the
wording of the rule guarantees that only subjects but not objects are affected by
it. In his system subjects are immediately dominated by S whereas objects are
immediately dominated by VP. Ross purposefully exempts non-sentential subjects
from his island constraints for empirical reasons. We will return to the status of
examples likes (2-a) in some detail in section 5.2, where we will conclude that such
examples are unlikely to involve genuine subextraction.
Chomsky (1973) gives up Ross?s restriction of applying the constraint to sen-
tential subjects only and formulates a generalized Subject Condition.
10
(4) No rule can involve X, Y in the structure
...X ...[? ...Y ...]...
where (a) ? is a subject phrase properly containing Y
and (b) Y is subjacent to X [my emphasis, JJ]
(Chomsky 1973:250, ex. (99))
Most subsequent work follows Chomsky in assuming a generalized view on subject
islands rather than Ross?s original generalization. In the following we will give a
brief overview of some of the major empirical generalizations and theoretical devel-
opments in the discussion of subject islands.
2.1.1.2 The Condition on Extraction Domains (CED)
Huang (1982) formulates the following principle:
(5) Condition on Extraction Domain (CED) (Huang 1982:505)
Aphrase A maybe extracted out of a domainB only if B is properlygoverned.
Essentially, Huang formalizes the descriptive generalization that there is a com-
plement/non-complement asymmetry for extractability (observed earlier by Cattell
1976). While objects allow extraction, subjects and adjuncts are impermeable for
movement. This is derived by a specific definition of (proper) government:
(6) Definition of (Proper) Government
(Huang 1982:156, adopted from Chomsky 1980:25)
? is governed1 by ? if ? is c-commanded by ? and no major category or
major category boundary appears between ? and ?.2
In a configuration such as (7) X0 (for X0 = lexical) properly governs its complement
YP as X0 c-commands YP and there is no intervener. The specifier ZP is not
1? is properly governed by ? if ? is governed by ? and if ? is lexical. (slightly modified from
Chomsky 1981:273)
2The second disjunct essentially defines barriers for proper government in terms of minimal c-
command, i.e. ?? ? such that ? asymmetrically c-commands ? and ? asymmetrically c-commands
?.
11
governed by X0 since it is not c-commanded by X0. Similarly, the adjunct WP is
not c-commanded by X0 and hence not governed.
(7) XP
WP XP
ZP X?
X0 YP
This yields the desired results in that only the complement is properly governed and
a phrase may be extracted out of it. Non-complements, i.e. specifiers and adjuncts,
are not properly governed and as such islands for extraction.
It is crucial that c-command be defined in terms of the first branching node
(Reinhardt 1981) and not in terms of the first maximal projection (m-command
in the sense of Aoun & Sportiche 1983). If the latter were adopted, the specifier
ZP would also be properly governed since it is dominated by the same maximal
projection XP as the governor.3
2.1.1.3 Freezing Effects
In order to understand Stepanov?s proposal we will have to briefly remind ourselves
of the notion of Freezing Effects. The term Freezing, as referring to a syntactic
node that is no longer permeable for extraction, was first used by Wexler & Culi-
cover (1981). Similar effects, however, were already hinted at by Ross (1974), who
compares the following examples:
3Note that in chapter 3 (p.159ff) Huang piggy-backs on this very distinction between the two
different command relations to account for the configurationality parameter. While fixed word
order languages follow the c-command definition of government, free word order languages adopt
m-command. This makes the interesting prediction, unnoted by Huang, that subjects in free word
order languages should not constitute islands. This sounds remarkably similar to the proposals
made by Stepanov we will discuss below. While it seems unclear whether we would want to
postulate a c- vs. m-command parameter, the descriptive generalization seems intriguing and
worth further investigation.
12
(8) a. [The Waco Post Office]j she will send [apicture oftj][to Inspector Smithers].
b. ??[The Waco Post Office]j she will send ti [to Inspector Smithers][a picture
of tj] .
He points out that ?it is more difficult to chop constituents from the shifted con-
stituent than it is to chop them from the unshifted one.? (Ross 1974: 103).
Wexler & Culicover (1981:542) formulate the Generalized Freezing Principle:
(9) A node is frozen if (i) its immediate structure is non-base or (ii) it has been
raised.
Condition (i) of this disjunction refers to phrase markers that could not have been
generated by base phrase structure rules, but have been formed as the result of a
transformation. This can be illustrated with example (8). After Heavy NP-shift of
a picture of the Waco Post Office in (8-b) the VP corresponds to a phrase structure
such as V PP NP. Assuming that VP ? PP NP is not a base rule of English, it
follows that no subconstituent of VP can be extracted out of it.
In some cases a transformation yields a phrase marker that could have also
been generated by a base rule. This is why Condition (ii) is needed, which can be
illustrated by the following example:
(10) a. Some people from Philadelphia greeted me.
b. Some people ti greeted me [from Philadelphia]i.
c. *[What city]j did you expect some people ti to greet you [from tj]i.
(Wexler & Culicover 1981:143)
In (10-b) the PP from Philadelphia has been rightward moved, yielding a phrase
marker of the type V NP PP. We know, however, that the rule VP ? V NP PP
is a base-rule of English, otherwise sentences like (8-a) could not be generated.
To account for the ungrammaticality of (10-c) we need the second condition of
13
the disjunction in (9), whereby ?raising? more broadly refers to any node that has
been subject to a transformation. Wexler & Culicover summarize their principle
by pointing out that only ?characteristic structures?, i.e. those phrase markers that
were constructed by base rules alone, may be affected by transformations. Structures
that have been ?distorted? by prior transformations may not be affected by further
syntactic rules.
2.1.2 Theoretical proposals
2.1.2.1 Multiple Spell-Out (MSO)
Uriagereka (1999) and Nunes & Uriagereka (2000) derive CED effects from general
derivational principles of a dynamic syntactic system. In particular, they do away
with the assumption that spell out (SO) applies at exactly one point of the deriva-
tion in favor of the assumption that spelling out structures to the interfaces is a
consequence of a general linearization requirement of mapping hierarchical struc-
ture into flat strings to satisfy PF legibility requirements. As a mapping algorithm
Kayne?s (1994) Linear Correspondence Axiom (LCA) is employed.
(11) Linear Correspondence Axiom (LCA)4
A lexical item ? precedes a lexical item ? iff ? asymmetrically c-commands
?.
Consider how the LCA applies in a structure such as (12) (terminal elements in
boldface):
4Note that this is the LCA as cited in Nunes & Uriagereka (2000:23), which holds in a SVO
language like English. While retaining Kayne?s basic insight that asymmetric c-command relations
are mapped into linear sequences, we will give up his assumption that SOV languages also have
an underlying SVO word order. This is discussed in some detail in section 3.6.1.
14
(12) VP
DP
friends PP
P
of
NP
Hermes
V?
V0
met
DP
friends of Kurt
Met asymmetrically c-commands (a-CC) its complement friends of Kurt and hence
also precedes it linearly5. It is easy to establish the precedence relations between
a verb and its complement, but what about the specifier? We are left with the set
of terminals {met,friends,of,Hermes}. Met is not in a-CC relation with any of
the other terminal elements and, consequently, the precedence relations cannot be
established. Essentially, this problem is not a peculiarity of the structure in (12)
but generalizes to all complex specifiers (and adjuncts). How can we resolve this
situation?
If we take the bottom-up derivation of our tree seriously, it follows that in
order to create a structure such as (12) we need to construct two different trees
in parallel in two separate workspaces. We are building the main spine and the
specifier in (13):
5We are abstracting away from the details of the internal structure of the DP for the moment.
Presumably the NP friends a-CCs the PP of Kurt, in which the preposition of a-CCs the NP
Kurt. This would derive the right precedence relations.
15
(13)
side workspace
DP
friends PP
P
of
NP
Hermes
main workspace
VP
V?
V0
met
DP
friends of Kurt
If SO were a single operation at the end of a derivation, we would end up with
an incomplete set of precedence relations and the derivation would not converge at
PF. Uriagereka (1999) chooses to drop this assumption in favor of a dynamic cyclic
application of SO. SO applies to the side workspace of (13) first, determining the
precedence relations internal to this workspace6. The structure is flattened out and
shipped off to the interfaces. As a result, only the label DP is available for further
syntactic operations and is plugged in its position in the main spine, as shown in
(14).
(14) VP
DP<friends,of,Hermes> V?
V0
met
DP
friends of Kurt
Uriagereka assumes that the DP has essentially been turned into a terminal element.
Since it a-CCsthe rest oftree, it precedes all other terminal elements inthe structure.
The precedence relations within the DP have also been established, i.e. ?friends? >
?of? > Hermes. He has to make the further assumption that all terminal elements
6For a discussion of how terminals that symmetrically c-command each other are spelled out
see section 3.6.1
16
dominated by DP, whose linearization information has already been shipped off to
PF, is retained in DP. We add the further requirment that, as DP precedes the rest
of the tree, all terminal elements dominated by DP also precede all other elements of
the tree. As a result, the set of precedence relations is complete and the derivation
can converge.
How does this relate to subject islands? If SO applies in a cyclical fashion for
reasons of linearization in the way just explicated, substructure of a phrase that has
already been spelled-out is no longer accessible for subsequent syntactic operations.
As a consequence, island effects aretheresult of cyclic SO. In our example athand we
can see that the elements internal to the DP friends of Hermes cannot be affected
by the syntax anymore, i.e. you cannot extract out of it. This directly predicts
classical subject island violations such as (15).
(15) *Whoi did friends of ti meet Kurt?
Extraction out of objects, on the other hand, is still possible. SO of the main spine,
which the object is part of, can be held back until the very end of the derivation.
As such, the substructure of the object is still available for syntactic operation and
MSO correctly rules in examples like (16).
(16) Who did Hermes meet friends of ti ?
How does the MSO fare with freezing effects? Every attempt to extract out of a
left branch will result in a significant decrease in acceptability, as we will see in
our studies below. To put it blatantly, once such an extraction has taken place,
that is that for the derivation. In section 3.2, however, we will see that pure CED
effects and freezing effects are additive. This seems to suggest that MSO alone is
not sufficient to give us full coverage of the data. We need an additional account of
freezing effects on top of it.
17
The situation is different if complements, which would normally be domains
transparent for extraction, become islands as a result of freezing. If we follow the
standard assumption that movement targets either a specifier or an adjoined posi-
tion, MSO gives us the desired result under some additional assumptions. Consider
the example in (17) (cf. Lasnik & Saito 1992):
(17) a. I think that you should read [articles about vowel harmony].
b. Whatj do you think that I should read articles about tj ?
c. I think that [articles about vowel harmony]i you should read ti.
d. ??Whatj do you think that [articles about tj]i you should read ti?
The object DP articles about vowel harmony, which normally allows subextraction
as shown in (17-b), is a frozen domain as a result of being moved, (17-d). Let us
see if this can be made to follow from MSO. We will assume that the DP articles
about what is topicalized to some adjunction position (say TP for concreteness). For
linearization purposes the DP needs to be spelled-out before it can be merged in
this position, as indicated by the italics in (18).
(18) [CP C [TP you [V P think [CP that [TP DP(articles, about, what) [TP you
should [V P read <articles about what> ]]]]]]]
When matrix C then probes for a wh-element, it cannot access the substructure
of the higher copy of articles about what, since it has already been spelled-out.7
It could, however, still access the lower copy at this point of the derivation and
movement of the lower what should be possible, incorrectly predicting that (17-d)
should be grammatical. Nunes & Uriagereka (2000) are aware of this problem and
suggest that no chains can be formed between two copies of a syntactic object if
7Norbert Hornstein (p.c.) points out that this is a non-trivial assumption, as it still needs to be
made more precise how and why SO renders structures in accessible to the syntactic component.
In a sense, the structure is stripped off the features relevant to the syntax. In other words, SO
converts a structure into a format that can only be read by PF but not by narrow syntax. It is
an interesting question of what exactly this mechanism is and how it could be formalized. This is
beyond the scope of this dissertation but clearly an interesting question for future research.
18
one copy has already been spelled-out while the other has not. For this reason they
stipulate that spell-out precedes copying, i.e. the lower copy needs to be spelled-out
before it can be moved to yield a licit chain, as shown in (19).
(19) [CP C [TP you [V P think [CP that [TP DP(articles, about, what) [TP you
should [V P read DP(articles, about, what) ]]]]]]]
Now, the lower copy of articles about what has already been spelled-out as well, when
matrix C tries to probe it and the derivation does not converge.8. Norbert Hornstein
(p.c.) points out that this might fall out from minimality, i.e. the higher copy is
structurally closer than the lower copy and hence the lower copy cannot be targeted
by C. Unfortunately, this is only true if we consider the copies in their entirety. The
higher copy (articles, about, what) indeed A-CCs the lower one. The wh-element
what itself, however, does not c-command out of its DP. Since this is presumably the
element C probes for, minimality does not solve the problem in any straightforward
way. Alternatively, it could be argued that the path of the higher copy of ?what?
to C is shorter than from the lower copy of ?what? to C. While that is true, it
presupposes that the higher copy of ?what? is still accessible for the computation
of the paths. This, however, would run counter the idea of cyclic SO since we
assume that SO renders a structure inaccessible to subsequent syntactic operations.
Computing paths quite plausibly requires the full pre-SO syntactic structure. There
does not seem to be any straightforward solution in terms of minimality.
Note that it is an empirical question whether Nunes & Uriagereka?s condition
on chain formation is necessary. This depends on the status of subextraction out
of subjects that originated in a complement position, i.e. unaccusatives, passives
and possibly complements of ECM predicates. Recall from the discussion in section
2.1.1.1 above that Ross judged extraction out of non-sentential passive subjects to
8See Nunes & Uriagereka (2000: 26ff.) for further discussion and Sheehan (2009) for an alter-
native solution.
19
be acceptable, as indicated by (2-a) repeated as (20):
(20) Of which cars were the hoods damaged by the explosion?
This position has recently undergone an renaissance with Chomsky?s (2008) advo-
cating for the well-formedness of the following examples:
(21) a. it was the CAR (not the TRUCK) of which [the (driver, picture) was
found]
b. of which car did they believe the (driver, picture) to have caused a
scandal
If these cases are indeed acceptable it seems like the structural position of the lower
copy does play a role after all, and Nunes & Uriagereka?s stipulation is not needed.
We will return to these cases in some detail in Chapter 5. Acceptability judgment
data will lead us to the conclusion that in English the structural position of the
lower copy is largely immaterial, i.e. a lower copy in the complement position does
not save the sentence from severe unacceptability.
2.1.2.2 Subject island effects as Freezing Effects
There have been recent proposals in the literature to do away with the CED com-
pletely (Stepanov 2001, 2007, Truswell 2007). The claim is that the CED is empir-
ically inaccurate and that extraction out of in-situ subjects is licit while extraction
out of adjuncts9 is ungrammatical across languages. Stepanov argues that sub-
ject island effects are reducible to freezing effects, i.e., subjects are not islands for
extraction per se but they become opaque as a result of being moved. That is
why languages like English, which obligatorily raise the subject to SpecTP, show
island effects while languages that allow for the subject to be in situ (e.g. German,
9There are reasons to doubt that the generalization about adjuncts is correct, see Cinque (1990,
Truswell (2007), Yoshida (2006) among others for counterexamples.
20
Japanese, Turkish) do not show such effects.
If Stepanov?s generalization is correct, this would have highly significant im-
plications for all theories that aim at deriving complement/non-complement asym-
metries (such as Chomsky 198610 or Uriagereka 1999). If such an asymmetry does
not exist, clearly there would not be any need for theories to account for it. Even
worse, such accounts would undergenerate since they predict that extraction out of
non-complements should be generally ruled out. Stepanov very explicitly tries to do
away with such theories. He proposes an ?eclectic account? of CED effects, reduc-
ing subject islands to freezing effects and treating adjunct islands by a completely
different mechanism. Concretely, he argues for accounting for subject island effects
with Takahashi?s (1994) Chain Uniformity condition, which bans the adjunction to
any link of non-trivial chains, i.e., previously moved constituents become opaque
for extraction out of them (= freezing). While Stepanov?s line of argument seems
compelling, there are reasons to question whether he provides a complete picture of
the data. Since the stakes are very high for a number of theoretical proposals, it is
of utmost theoretical importance to get a clearer picture of the facts.
2.2 The inadequacy of informal judgments
Before we delve into the details of our experiments on CED effects crosslinguistically,
a few methodological points need to be clarified. The core of this thesis is formed by
data of controlled acceptability judgment studies. While more and more linguists
employ acceptability judgment studies, and their usefulness has become less and
less a point of contention, I still think it is worthwhile to dedicate some space to
why such experiments can give us invaluable insight into certain empirical domains.
This section will focus on general points. Arguments for the need for more reliable
10For reasons of space we will not summarize Chomsky?s Barriers framework. The reader is
referred to Chomsky (1986) or a more digestible version in Lasnik & Saito (1992:69-75).
21
data in the domain of CED effects in particular will be given in later chapters.
2.2.1 De iudiciis non est disputandum?
I know of no linguist who has never stumbled across an example sentence in a
paper on his native language that prompted him to exclaim one of the following:
Really? That?s supposed to be grammatical? or No way! That sounds totally fine!11
Sometimes this very example happens to be a crucial data point in the discussion,
and you might start wondering: I am native speaker, the author is a native speaker,
or if not he or she claims to have consulted with native speakers. And still we do
not seem to have the same opinion about this judgment. What is going on?
What causes these disagreements concerning judgments? One logical possibil-
ity is clearly that we are faced with different idiolects. After all, it is I-language we
are interested in, not E-language or vague and virtually indefinable notions such as
English, German or Japanese. If we disagree on a specific judgment, our idiolects
differ, and there is nothing else we can do. This is a logically coherent position and
a very tempting one, but one that is a non-starter if we are interested in serious
investigation of a given phenomenon. Here is why:
We cannot directly observe people?s abstract mental representations of their
grammar. We can only indirectly access them by eliciting well-formedness judgments
onspecific outputs ofthisI-language, i.e. concrete example sentences. Ajudgment of
this sort is by definition subjective and as such not amenable to objective scientific
inquiry. Of course any native speaker can insist on a particular judgment, and
it is logically impossible to prove him or her wrong. This rather stubborn and
dogmatic position is what Featherston (2007, p. 279) quite appropriately dubs
the ?my idiolect? gambit. With it, any further discussion is scotched. Any chance
of drawing generalizations based on observable trends within a larger number of
11For semantics papers we might add: Of course this reading is available! What are you talking
about?
22
judgments and speakers is forfeited. Playing the ?my idiolect? gambit comes at the
price of unfalsifiability. In short, we are in danger of leaving the realm of science.
Linguistic judgments are not matters of taste in the sense that preferences in
food or music are. There is no clear answer to whether Bob Dylan?s latest album is
good or bad. I am a big admirer of his, but I believe it is somewhat overrated, an
opinion for which many Dylan fans and music critics around the world would throw
tomatoes at me. And that is not a problem. Tastes differ. It does not, however,
seem particularly desirable if linguists were to turn to throwing tomatoes whenever
they disagree about a judgment. Judgments in linguistics are not a matter of taste
in the same sense as opinions about records. Unlike taste, there is indeed arguing
about acceptability judgments. If we find ourselves in disagreement, it is not good
enough to shrug our shoulders and conclude that this is an unsolvable dispute. We
can try harder, we can control for potential interfering factors, we can ask a larger
number of speakers under the same conditions, we can gather judgments about
multiple lexicalizations of a certain construction, and we can resort to well-known
statistical tests to separate the noise from the meaningful signal we are interested
in.
Perhaps, after we have controlled for all these factors, it in fact turns out
that our judgments reflect dialectal differences of some sort. Great! We now have
settled the dispute, and it is indeed the case that we are both right. We both win!
How often does that happen? We can now proceed to worrying about the source of
this variation, if it holds systematically for an entire construction or just for a few
isolated examples and if it correlates with other differences between our grammars.
Importantly, however, this is only the second step after we have convincingly and
thoroughly established that such a difference in fact exists.
23
2.2.2 Operational procedures
The question of how to obtain reliable data about linguistic competence has been
a matter of concern since the earliest days of generative grammar. Informal intro-
spection was accepted as the technique of choice, not so much out of conviction but
rather because of a lack of alternative methods. In Aspects, Chomsky very explicitly
discusses the limitations of informal judgments and anticipates that there might be
better techniques in the future:12
In brief, it is unfortunately the case that no adequate formalizable tech-
niques are known for obtaining reliable information concerning the facts
of linguistic structure (nor is this particularly surprising). There are,
in other words, very few reliable experimental or data-processing pro-
cedures for obtaining significant information concerning the linguistic
intuition of the native speaker. [...] If operational procedures were avail-
able that met this test, we might be justified in relying on their results
in unclear and difficult cases. This remains a hope for the future rather
than a present reality, however. [my emphasis, JJ]
[Chomsky 1965, p. 19]
Fortunately, the field has made significant progress since the sixties, and more re-
liable experimental procedures of the kind Chomsky anticipated for the future can
now be called upon in the case of doubtful judgments. Bard et. al (1996), Sch?utze
1996, Keller (2000), Sprouse (2007) and Featherston (2007) are only a few instances
of the ever growing popularity of experimental data elicitation.13 A number of
requirements and desiderata for improved and experimentally more sound data col-
lection have been proposed. I will only list the requirements here without much
comment about their motivation or their usefulness. The reader is referred to the
meticulously thorough discussion in Sch?utze (1996) and Featherston?s (2007) pro-
grammatic paper:
12Thanks to N. Hornstein for bringing my awareness to this passage.
13Mind you, these techniques obviously still only collect people?s judgments about the acceptabil-
ity of a certain string and as such only indirectly shed light on the underlying mental grammar. We
still have not invented machines to directly measure people?s intuitions, and it is highly doubtful
if such techniques will ever exist or if their existence is even possible.
24
? The items are presented as controlled conditions, i.e. sentences only vary
in the one factor that is being tested while everything else is held constant
(minimal pairs).
? Speakers are presented with several lexicalizations per condition to test
identical constructions using different lexical items. This minimizes the chance
of mistaking a purely lexical for a structural effect.
? The test items are interspersed with balanced fillers, i.e. speakers see fillers
of all levels of acceptability (from monoclausal wh-questions to CSC viola-
tions).
? The stimuli are presented in a Latin square design to guarantee that po-
tential noise is spread across conditions in order not to bias one condition
disproportionally.14
? The test items and the fillers are pseudo-randomized, presenting the items
in a random order while making sure that fillers and test sentences alternate.
? Only ?naive? participants are tested, i.e. speakers that have had no prior
training in formal linguistics.15
14In a Latin square design the first lexicalization of the first condition is assigned to the first list
of stimuli, the second lexicalization of the first condition is assigned to the second list of stimuli,
the third lexicalization of the first condition is assigned to the third list of stimuli. Then the second
lexicalization of the second condition is added to the first list, the third lexicalization of the second
condition is added to the second list etc. until all lexicalization of all conditions are assigned to a
list. This is illustrated in the following table representing a Latin square design for an experiment
with four conditions (S = list of stimuli, C = condition, L = lexicalization):
i.
S1 S2 S3 S4
C1 L1 L2 L3 L4
C2 L2 L3 L4 L1
C3 L3 L4 L1 L2
C4 L4 L1 L2 L3
15It is a desideratum in linguistic experiments to exclude trained linguists from the pool of
25
? Participants are given detailed instructions to make sure they understand
what is being tested (descriptive vs. prescriptive judgments, grammatical
acceptability vs. pragmatic plausibility etc.)
? Participants are given test trials to get accommodated with the task.
These are all criteria that are relatively straightforward to incorporate in the data
gathering process. Running an experiment, however, is still more cumbersome than
engaging in what is sometimes affectionately called ?armchair linguistics?, or asking
your colleague next door to engage in ?armchair linguistics? with you. So is there
a reason to frantically jump out of your armchair and anxiously inquire: I have
to do all of this every time I want to report a judgment? There is no reason for
concern. We do not need to crack every syntactic nut with a sledgehammer. Some
examples are just as obviously good as others are bad. A classic case is subject-verb
agreement in English:
(22) a. John is here.
b. *John are here.
It would undoubtedly be a waste of time and research money to run a full-fledged
acceptability judgment study if all you are trying to show is that there is contrast
between (22-a) and (22-b). Armchair linguistics is clearly the method of choice
here.16 And only looking at clear cases of this sort can indeed give us considerable
participants. This is done to prevent speakers from rating judgments based on their theoretical
beliefs rather than their introspective intuitions (see Sch?utze 1996:113ff. for discussion and a
summary of experimental studies indicating that this is in fact true). That linguists are in fact
biased and hence worse judgers has recently been challenged (see Culbertson and Gross 2009 for
discussion). For the time being I will follow the current practice of excluding linguists from the
experiments, not because I believe them to be prone to purposefully giving wrong judgments to
provide evidence for their or someone else?s theory, but mainly because the theoretical beliefs you
hold literally change the reality of your judgment. Testing only people that do not have any explicit
beliefs about a given structure is a way to avoid this Constructivist trap.
16Even with something seemingly as straightforward as subject-verb agreement in English things
can get tricky: The sentence The key to the cabinets are on the table. sounds significantly more
26
mileage and might be sufficient in certain domains (see Phillips 2009 and Featherston
to appear for essentially this conclusion).
Crucially, more formal methods of data collection do by no means replace
proper linguistic analyses. Luckily, this is not a question of either or. A reasonable
balance between data and theory is desirable. And there are many areas where less
straightforwardfacts have adirect bearing onour theory. Andclearly, as Featherston
(2007) points out repeatedly, our theory is only as good as our data. We will see
throughout this thesis that islands in general and CED effects in particular are a
domain where these kinds of studies can be usefully applied.
2.2.3 Arrogance and modesty
Why then is there still a reluctance in employing experimental techniques in theo-
retical linguistics? Why do recent papers, published in prestigious journals, often
still use data to support their theories that has to be classified somewhere between
highly controversial and blatantly wrong? (we will return to numerous examples
of such cases in the next chapters). Haider (2009) addresses these questions in a
trademark humorous fashion. He invokes the 19th century psychologist Wilhelm
Wundt (1845-1920) who quite eloquently attributes the theoreticists? opposition to
experiments to two seemingly contrary forces, arrogance and modesty.
Die eine Eigenschaft ist der Hochmuth. Es gibt ja immer noch einige
Leute, die das Experimentieren f?ur eine banausische Kunst halten, mit
der man sich nicht befassen d?urfe, wenn man nicht des Privilegiums, im
Aether des reinen Gedanken zu hausens, verlustig gehen wolle. [...] Die
andere Eigenschaft ist die falsche Bescheidenheit. Jede Kunst scheint in
der Regel dem, der sie nicht versteht, viel schwerer als sie wirklich ist.
[One property is arrogance. There are still some people who consider
experimenting a philistine art, which one should not deal with, if one
does not want to risk losing the privilege of residing in the pure ether
acceptable to speakers than (22-b), at least on a first parse. Here, more controlled experimental
work is clearly desirable and has in fact been done (see Phillips et al. to appear and the references
therein for discussion).
27
of thoughts. The other property is mistaken modesty. Every art usually
tends to appear to be more difficult than it really is to those who do not
understand it., translation by Haider (2009)]
[Wundt 1888, p.292-309]
Wundt is widely regarded as the founding father of experimental psychology. Haider
points out that Wundt was highly influential in transforming the field of psychology
from a discipline largely interested in ?the pure ether of thoughts? to a hard science
comparable to physics or chemistry. Haider goes on to argue that linguistics has
not made this transition yet but is only currently undergoing a transformation from
the dark ages of alchemy to the mature science of chemistry. Linguistics, he claims,
still contains ?quite some phlogiston-theories?.
Haider?s reckoning with the current state of the field of linguistics certainly
needs to be taken with a grain of salt. But just like other comical exaggerations,
his discussion contains certain elements of truth, which most experimentalists are
likely to have experienced themselves when presenting studies on controversial data
to a group of theoretical scholars. The bottom line of Haider?s discussion is: ?if
linguistics wants to be respected as a branch of (cognitive) science, it has to accept
and apply the empirical standards of (cognitive) science?. While this might be
somewhat blown out of proportion for dramatic effects, a little more care when
collecting data certaintly cannot hurt. At the same time, we must not forget that
the data is never free of theory, nor should it be. Our theoretical constructs point
us to look at specific pieces of data, which in turn might cause us to modify our
theoretical constructs. Or as Kant put it so pointedly in his Critique of Pure Reason:
Gedanken ohne Inhalt sind leer, Anschauungen ohne Begriffe sind blind.
[Thoughts without content are empty, intuitions without concepts are
blind]17.
17Thanks to Norbert Hornstein for providing me with this Kant quote
28
2.2.4 Judgment = Grammar + X
To say that the relation between acceptability and grammaticality is non-trivial and
poorly understood is one of the biggest euphemisms in linguistics. We have already
mentioned that we can only gather acceptability judgments by introspection of our
own native speaker knowledge or by eliciting data from other native speakers. The
term ?grammaticality judgment? is meaningless. There are no judgments about the
grammaticality of a given sentence. We might state that sentence S is grammatical,
by which we mean that our grammatical theory allows S to be generated. But
we do not have any intuitions about the grammaticality of a sentence. Sometimes
we can immediately point to the grammatical constraint which we assume to have
caused our perceived decrease in acceptability. If the violation is blatant enough,
non-linguists might be able to do so too. This still does not imply that we have
intuitions about the grammaticality of a sentence. It only means that our intuitions
tell us that sentence S is unacceptable, and post hoc analytic reasoning sometimes
allows us to identify a grammatical principle responsible for this degredation.
The problem we face whenever we gather an acceptability judgment and try
to make claims about the underlying grammar is the following:
(23) Judgment = Grammar + X
What is the X? This is ultimately a question that cannot be decided on the basis
of the data alone but requires certain theoretical assumptions. It is a logically
coherent position, and the one assumed by a large portion of the field, that the
grammar is binary and all gradience we find in our judgments comes from the X-
factor. This leads many theoretical linguists to tacitly adopt a binary view on
grammatical constraints. A direct consequence of this position is the assumption
that if sentence A violates constraint X, it is ungrammatical. If there is an instance
of sentence A that some speakers find acceptable, constraint X does not exist in this
29
language or is wrongly understood. The radically binary view on grammaticality
is expressed pointedly by Haider (1993:159), who argues that in order to disprove
the correctness a of grammatical principle a single example is sufficient (original
emphasis)18
This statement explicitly presupposes viewing grammaticality in terms of ab-
solute categoricity, i.e. a sentence is either grammatical or not. Since the earliest
days of generative grammar, however, this view has been known to be nothing more
than a useful abstraction of the facts. In Syntactic Structures, Chomsky (1957)
points to the need for developing a ?notion of degree of grammaticalness?. In As-
pects (Chomsky 1965), he emphasizes that ?like acceptability, grammaticalness is,
no doubt, a matter of degree? and later he reiterates that ?an adequate linguistic the-
ory will have to recognize degrees of grammaticalness? (Chomsky 1975). Note that
Chomsky consistently talks about gradient grammaticalness and not just gradient
acceptability.
Despite being very explicit that grammaticality is gradient, Chomsky, andwith
him the majority of theoretical syntacticans, have worked under the abstraction
of binary categoricity. However, when working with real data, they very often
encountered cases that would not nicely fit into one of those two categories, and
they saw the need for intermediate judgments. This resulted in a quite creative
semiotic system, including such notations as: *, ?, **, ??, *?, #, %, ?, ?, ?, ??, etc.
If grammaticality comes in a continuous scale but we consistently try to
squeeze it into a binary distinction, it does not come as a surprise that we find
contradictory judgments in the literature. It is quite absurd to give people a grey
square and asked them whether it is ?black? or ?white?. Some people will say it is
black, others that it is white. Does that mean people?s perception of color differs?
This seems like a highly unlikely conclusion. What seems to be happening is that
18Given Haider?s recent fervent advocacyof experimental methods in theoretical syntax we might
assume that he has slightly changed his position. But this is only speculation.
30
we are asking a misleading question.19
Returning to (23), we have to stress that the question of whether the gradience
found for a given sentence comes from the grammar or from the X-factor cannot be
decided on purely empirical grounds. Every datum is interpreted by a researcher
who is forced to make a theoretical decision. Of course, we can locate certain
factors we know are part of X and control for them as meticulously as possible.
However, given that our understanding of X and how it interacts with the grammar
is very limited, there is always a point where we have to make assumptions about
where gradience is actually reflective of the grammar itself. This holds true for the
experiments discussed in this thesis as well. I will try to be as explicit as possible
about these assumptions and justify them as well as I can.
It needs to be emphasized that it is not within the scope of this dissertation
to provide any detailled theory of how gradient grammaticality could be modelled.
The reader is referred to Keller (2000, chapters 6 and 7) for very insightful discus-
sion. Furthermore, I do not have much to add to big picture questions such as the
epistemological nature of judgments or the biological reality of the notion of gram-
maticality. The goal - more modest but potentially more realistic - is to employ
more fine-grained experimental techniques that can capture previously unnoticed
differences in acceptability. These differences in acceptability are argued to be re-
flective of underlying differences in grammaticality. These differences allow us to
provide evidence for one theory over another, and as such bring us closer to a better
understanding of how the language faculty operates.
19In fact it turns out that gradience even manifests itself in binary judgment task if enough data
is collected. Goodall et al. (2010) compared yes/no with Magnitude Estimation (ME) and 7-point
scale tasks and found that all three methods led to statistically non-significantly different results.
Yes/no and 7-point scale tasks even had the advantage of producing less noise than ME. Their
studies, however, only investigate very strong contrast. It would be worthwhile replicating these
comparisons with more subtle differences.
31
2.2.4.1 Capturing gradience
Experimental studies following the methodological protocol discussed in the last
section can incorporate gradience in a number of ways. The technique of choice has
often been Magnitude Estimation (ME) (e.g. Bard et al. 1996, Keller 2000). In
ME speakers are given a base line sentence and are asked to assign this item an
arbitrary numerical score, say 100. Then they judge the test items with respect to
this baseline or reference sentence, e.g. base line items A receives a score of 100,
item B sounds half as acceptable and gets 50, item C is somewhere in between A
and B and gets 75 etc. ME has been argued to not force a scale on people and be
in principle open-ended (see Sprouse 2007, p.11ff for detailed discussion).
ME has been praised to the skies as the best technique for gathering accept-
ability judgments, to the extent that ME studies became almost a synonymous term
with controlled acceptability judgment experiments. In ME no scale is imposed upon
the speakers so that even the most subtle difference can be encoded. The scale is in
principle open ended and ME data is interval data and can be analyzed using para-
metric statistics. This entails that the intervals between the data are meaningful
and that the observations are normally distributed.
Controlled experiments and ME, however, are not mutually inclusive. They
are two methodological aspects that have to be kept separate. There can be very
poorly controlled studies using ME just as there can be very carefully designed
experiments employing techniques such as 5 or 7-point Likert scales or even binary
Yes/No tasks. As a matter of fact, there have been a number of meta-studies
(Sprouse 2007, Weskott & Fanselow 2008, Murphy & Vogel 2008, Goodall et al.
2010) all pointing towards the conclusion that ME does not yield better data than
other techniques. Weskott & Fanselow (2008, p.431) argue that ?not only do Likert-
scale judgments provide the same amount of information about a given empirical
hypothesis, but also that the inherent variability of ME judgments makes them more
32
susceptible to the production of spurious variance.?
Sprouse points to a number of potentially severe problems. First, he finds
that speakers impose categoricity on the continuous scale. As a result, ME is not
immune to floor and ceiling effects, as it is often praised for. Second, he finds that
the choice of the reference not only effects the absolute values participants assign
to each item but even the ranking between them. In other words, item A is judged
more acceptable then item B if X is the reference sentence, but B is more acceptable
than A if Y is the reference sentence. If this finding is confirmed in other studies,
this would be a potentially detrimental problem for ME. The experimenter could,
consciously or unconsciously, bias the results in this way. Needless to say, this
would not be an improvement compared to informal introspection data. It might be
even worse since skewed data would present itself in the guise of hard experimental
results.
For the above reasons all experiments in this thesis employed a 7-point Likert
scale instead of ME. We will follow the methodological protocol outlined in the
previous section. Further details about the methodology of each experiment will be
given in the respective chapters. All stimuli used can be found in the Appendix.
2.2.5 Christmas trees
In conclusion of this chapter let me point to another source of disagreements on
specific informal judgments, which can be found over and over again in the theoret-
ical literature. To illustrate, let us consider an anecdotal analogy. Last Christmas I
went to see the Christmas tree in front of Rockefeller Center for the first time in my
life. I was very excited. I had seen it on TV and always pictured it as gigantically
huge. When I saw it in reality I was a bit disappointed. ?This is not as big as
I imagined. I have seen bigger!? Returning to Washington DC, I was passing the
White House and saw the Christmas tree there. ?Now, that is a big Christmas tree!?
33
Why are people only talking about the one in front of Rockefeller Center?? In my
head images similar to the ones in Figure 2.1 were popping up.
Figure 2.1: Christmas Trees in front of the White House and Rockefeller Center
It took me a minute before I made the relatively trivial realization that the
White House Christmas Tree is next to a 58-foot (17.7m) tall building, while the
Rockefeller Center Christmas is right in front of a 872-foot (266 m) tall building
with 70 floors. I looked up the sizes of the trees later and it turns out that the
Rockefeller Center tree is in fact 58-foot (17.7m) higher than the White House Tree
(76-foot vs. 18-foot).
Why am I talking about Christmas Trees in a dissertation on syntax? The
optical illusion created by the relative placement of an object was strongly reminis-
cent of what can be found in many theoretical papers. The author is presenting a
perceived contrast between sentence A and B, and immediately concludes that A is
grammatical and B is ungrammatical. A different author notices a contrast between
B and C and concludes that B is grammatical while C is ungrammatical. Now we
have two authors, the one claims that B is ungrammatical, the other one claims that
B is grammatical? Who is right?
34
The point is that both are right in that they have noticed a contrast in accept-
ability between two examples. They might both be wrong at the same time in the
conclusions they draw. Presenting a pair-wise contrast is often not good enough,
especially in domains with intermediate judgments. The full paradigm is needed.
Let me illustrate with an example from German. We will return to these cases in
the following chapter.
(24) a. *Was
what
haben
have
f?ur
for
Leute
people
deine
your
Mutter
mother
besucht?
visited
?What kind of people visited your mother??
b. Was
what
hast
have
du
you
in
in
Italien
Italy
f?ur
for
Museen
muesums
besucht?
visited
?What sort of museums did you visit in Italy??
[den Besten 1985]
(25) a. Was
What
haben
have
denn
PRT
f?ur
for
Ameisen
ants
einen
a
Postbeamten
postman
gebissen?
bitten
?What kind of ants bit the postman??
b. *Was
What
haben
have
f?ur
for
Ameisen
ants
denn
PRT
einen
a
Postbeamten
postman
gebissen?
bitten
?What kind of ants bit the postman??
[Diesing 1990]
(24-a) is judged unacceptable by den Besten (1985), while (25-a), which is virtually
identical in all relevant respects, is judged acceptable by Diesing (1990).20 In both
cases was-f?ur split out of the subject of an unergative has taken place. What
differs, however, is the two authors? frame of reference. Diesing contrasts it with
the extraction out of a derived subject in (25-b), while den Besten contrasts it with
extraction out of an object in (24-b). This fits exactly the template outlined above.
(24-a) sounds relatively worse compared to (24-b), while (25-a) sounds relatively
20One difference is the presence of an adverbial particle in (25-a). This forces a construal of
the subject in vP and not TP. This structural analysis, however, should also be possible, yet not
required, in (24-a). We will return to these issues in the next chapter.
35
better compared to (25-b). If you are forced to make a binary choice, it is no
surprise that Diesing and den Besten reach the exact opposite conclusions. Both
authors? point of view is skewed because neither considers the full data set to realize
that the status of B, extraction out of external subjects in this case, is in fact
intermediate. This is what we will find confirmed in a number of studies, discussed
in the following chapters.
36
Chapter 3
NP-Subextraction in German
3.1 Introduction
In this chapter we will investigate subextraction out of NPs in German. To this end,
we will scrutinize the was-f?ur split construction. We are primarily interested in how
the origin of the extraction site influences the overall acceptability of the sentence.
In particular, we will vary whether the extraction took place out of an internal
argument, i.e. objects, unaccusative and passive subjects, or out of an external
argument, i.e. intransitive and transitive unergative subjects. Furthermore, we will
investigate the role of freezing effects, i.e. whether the extraction originated from
a moved or an in-situ domain. This will allow us to shed light on a controversial
and largely unsettled issue in the theoretical literature on German and to test the
claims made by Stepanov (2007), discussed in section 2.1.2.2.
German is one of the languages claimed to allow extraction out of in-situ
subjects (Haider 1983, 1993, Diesing 1992 among others, see section 3.1.3 for a
review of the literature).was-f?ur split is a very useful construction in that it allows us
to tease apart CED effects from freezing effects. We will see that all the experiments
discussed in this chapter converge on the conclusion that CED effects need to be
kept separate from freezing effects. German shows complement/non-complement
asymmetries even in unmoved domains, while extraction out of moved constituents
incurs further cost.
37
3.1.1 What?s a subject?
Before we delve into the discussion of the structural restrictions on subextraction
in German in general and the (non)transparency of subjects in particular, some
terminological clarifications of the terms ?subject? and ?subject position? are in order.
The term ?subject? is drawn from traditional descriptive grammar and is notoriously
ill-defined.1
The term ?subject? is not normally a primitive or formally defined notion in
generative grammar. It has been, however, commonly used as a descriptive term to
refer to a number of closely related yet often not identical concepts (see McCloskey
1997 for an overview of the notion ?subject? in generative grammar). We will concern
ourselves with the notion of subject primarily as a syntactic entity and we will only
be interested it its semantic aspects to the extent that they affect its structural
position.
It is important to note that we do not expect UG to impose constraints on
inhomogeneous and ultimately undefinable concepts such as subjects. Ross?s (1967)
original formulation of the Sentential Subject Constraint was already stated in terms
of a constraint on an abstract structural description:
(1) The Sentential Subject Constraint (SSC)
No element dominated by an S may be moved out of that S if that node S is
dominated by an NP which itself is immediately dominated by S.
Ross avoids the term ?subject? in the definition of his constraint but the wording
of the rule guarantees that only subjects and not objects are affected by it. In his
system subjects are immediately dominated by S whereas objects are immediately
dominated by VP.
1It goes back at least to Aristotle?s Organon who coins the term uni1F51?????????? or ?the underlying
thing? (lat.: subiectum), which corresponds to an entity that can be predicated over (for a history
of the term see C.J.F. Williams 1985).
38
While the term ?subject? was (and still is) widely used for expository purposes,
we expect UG to operate on more abstract and more general notions. Specifically,
if Uriagereka (1999) is on the right track it seems that what could descriptively be
termed complement/non-complement asymmetries follows from the structure build-
ing mechanism of phrase markers. In essence, extraction is licit only out of elements
that have been assembled as part of the main spine. Such a theory is quite ap-
pealing aesthetically and from the point of parsimony. But of course it can only be
maintained if it covers the empirical facts in an appropriate and satisfactory fashion.
It is a central claim of this dissertation that the generalizaton according to which
complements are the preferred extraction domains is empirically tenable.
3.1.2 Subject positions in German
3.1.2.1 SpecCP
The term ?subject position? in German is used in the literature somewhat confusingly
to refer to a number of different positions. For our purposes it is crucial to make
precise what we mean by this term. Importantly, we do not refer to the first position
of V2 declarative clauses. The latter is traditionally identified as a topic position
and linked with SpecCP in generative analyses (cf. den Besten 1977/1983).2 While
the German subject can occupy this topic position - and in fact does so in the
information-structurally unmarked case - it need not do so and can remain in the
Mittelfeld, i.e. following the finite verb. I will refer to the sentence initial position in
V2 clauses as ?topic position? or SpecCP (cf. Svenonius 2001, 213ff. for discussion
of the topic position in Germanic).
2I am agnostic as to whether the C-domain in German is more complex along the lines suggested
for Italian by Rizzi (1997). If such a view were adopted the topic position would be identified with
SpecTopP.
39
3.1.2.2 SpecvP and SpecTP
Furthermore, we will assume that German hastwo subject positions in the Mittelfeld,
an in-situ position internal to vP and a derived position in SpecTP (as argued for
by Webelhuth 1989, Grewendorf 1989, Diesing 1992, M?uller 1999, 2010, Wurmbrand
2004 and Stepanov 2007). Let us consider two arguments for the existence of these
positions. The reader is referred to the literature for additional evidence.
Indefinite subjects can optionally raise to TP or stay in-situ. Definites, on the
other hand, obligatorily raise to TP in the unmarked case. Consider the following
examples (subjects in boldface, capitals indicate prosodic stress):
(2) a. Es
it
hat
has
ein Lama
a lama
doch
indeed
tats?achlich
in.fact
in
in
meinem
my
Garten
garden
?ubernachtet.
spend.the.night
?(Believe it or not) A lama spent the night in my backyard.?
b. Es
it
hat
has
doch
indeed
tats?achlich
in.fact
ein Lama
a lama
in
in
meinem
my
Garten
garden
?ubernachtet.
spend.the.night
?(Believe it or not) A lama spent the night in my backyard.?
c. Es
it
hat
has
das Lama
the lama
doch
indeed
tats?achlich
in.fact
in
in
meinem
my
Garten
garden
?ubernachtet.
spend.the.night
?(Believe it or not) The lama spent the night in my backyard.?
d. #Es
it
hat
has
doch
indeed
tats?achlich
in.fact
das LAMA
the lama
in
in
meinem
my
Garten
garden
?ubernachtet.
spend.the.night
?(Believe it or not) It was the lama that spent the night in my backyard.?
In these examples the adverbial particles doch tats?achlich are used to demarcate
the VP-boundary (as proposed by Webelhuth 1989, Diesing 1992 among others).
German abounds with modal particles of this sort, which are notoriously difficult to
translate. Doch tats?achlich adds a flavor of surprise and dismay to the proposition.
40
I am not sure how well my English translation covers this meaning but it seems to
go in the right direction.
(2-a) and (2-b) show that the indefinite subject ein Lama can appear either
in its VP-internal position or raise above the adverbial particles to TP. It should
be noted that the contrast between (2-c), where the definite subject has raised, and
(2-d), where the definite subject follows the adverbial particles, is not a contrast in
terms of grammaticality. (2-d) is not ungrammatical but has a marked word order,
as indicated by the # sign, and is only felicitous in particular pragmatic contexts.
(2-a), (2-b) and (2-c) can all be uttered in an out-of-the blue context. (2-d), however,
is only felicitous with prosodic stress on das Lama, indicated by the capitals. The
fact that some spending the night-event took place is now background information
and the focus is shifted to the fact that a lama is agent of this event opposed to,
e.g., a wildebeest. No such restrictions apply to the indefinites in (2-a) and (2-b).
I take this marked reading to be reflective of some sort of non-canonical operation
taking place in (2-d). The movement of the indefinite in (2-a) and (2-b) is analyzed
as optional A-movement (as argued for by Wurmbrand 2004) and does not have any
effect on the information-structural properties of the sentence.
Further evidence for two Mittelfeld-subject positions comes from VP-topicali-
zation. As it is well known, the first position (or Vorfeld) of the German declarative
clause - normally associated with the SpecCP position - can be filled by any XP,
including VP. It has been observed (Grewendorf 1989) that indefinite subjects can
front with the VP while definite subjects cannot.
41
(3) a. [Ein
a
Bauchredner
ventriloquist
aufgetreten]
performed
ist
is
bei
at
der
the
LSA
LSA
noch
never
nie.
before
?As far as ventriloquists are concerned, none has ever performed at the
LSA.?
b. ?[Ein
a
Bauchredner
ventriloquist
einen
a
Vortrag
talk
gehalten]
given
hat
has
bei
at
der
the
LSA
LSA
noch
never
nie.
before
?As far as ventriloquists are concerned, none has ever given a talk at the
LSA.?
c. *[Der
a
Bauchredner
ventriloquist
aufgetreten]
performed
ist
is
bei
at
der
the
LSA
LSA
noch
never
nie.
before
?As far as the ventriloquist is concerned, he has never performed at the
LSA.?
d. *[Der
a
Bauchredner
ventriloquist
einen
a
Vortrag
talk
gehalten]
given
hat
has
bei
at
der
the
LSA
LSA
noch
never
nie.
before
?As far a the ventriloquist is concerned, he has never given a talk at the
LSA.?
It should be pointed out, especially in a dissertation like this one, that the judgments
could be clearer with respect to (3-b) (see Hankamer & Schoenfeld 2005 for a semi-
formal questionnaire based study on data of this kind). The unaccusative subject
in (3-a) fronts the most easily, whereas the unergative subject in (3-b) is a little
degraded.3 Both, however, and that is the crucial point, are quite significantly better
than (3-c) and (3-d) where the subject is a definite (the unaccusative/unergative
contrast seems to go away as well).4
3If a contrast between (3-a) and (3-b) indeed exists and assuming that unergative subjects are
in SpecvP , it could be proposed that VP fronts more easily than vP . This, however, would not
fare well with Wurmbrand?s (2004) account, summarized in the main text below. This question
has to be left open until a proper investigation of the data is conducted.
4It should be noted that the definiteness restriction does not hold for objects, as (i) and (ii)
show. This is evidence that the fronting restriction is not reducible to a general constraint on
fronting VPs containing definite DPs .
i. Einen
a
Plenarvortrag
plenary.talk
gehalten
given
hat
has
der
the
Bauchredner
ventriloquist
bei
at
der
the
LSA
LSA
noch
never
nie.
before
42
Wurmbrand (2004) accounts for this contrast by assuming that only vP can
front but not TP. She relates this proposal with Abel?s (2003) assumption that
complements of phase heads cannot move. As a result, VP and TP cannot front,
only vP can, as illustrated in (4).5
(4) Wurmbrand?s (2004) analysis of German vP -fronting
a. CP 
  2 
              C' 
           2 
        C            TP 
     Aux       2 
               DP           T' 
          (def) SUBJ 2 
                        vP           T 
                    2       tAUX 
                 tDP           v' 
                           2 
                       VP            v 
                    5 
   * 
b.  CP 
                  2 
         C' 
                          2 
                       C            TP 
                     Aux      2 
                                              T' 
                                         2 
                               vP                   T 
                           2                 tAUX 
                      DP             v' 
              (indef) SUBJ  2 
                               DP          v' 
                   (def) DIR OBJ 2 
                                   VP             v 
                                2 
             gid1            tDP           V 
This account directly captures the data in (3). Definite subjects obligatorily raise to
SpecTP and are no longer part of vP when the fronting operation applies. Indefinite
subjects, on the other hand, can optionally stay in-situ and can be part of the fronted
constituent.
These are only two among many others arguments presented in favor of two
ii. Den
the
Plenarvortrag
plenary.talk
gehalten
given
hat
has
der
the
Bauchredner
ventriloquist
bei
at
der
the
LSA
LSA
noch
never
nie.
before
?As for a/the plenary talk, the ventriloquist never gave one/it at the LSA.?
5This analysis seems especially plausible with respect to TP. TP is already the complement of
C, so movement of TP to SpecCP would amount to re-merging a phrase with the same head and
thus blatantly violate anti-locality.
43
Mittelfeld subject positions in German (see Grewendorf 1989 for additional syntac-
tic arguments and Diesing 1990, 1992 for semantic arguments). For the sake of
completeness, we will briefly mention a slightly diverging point of view, advocated
by Haider (1993). He claims that German has eine einzige und VP-interne Sub-
jektsposition (?only one and VP-internal subject position?, my translation JJ). The
details of the wording here are important. Haider is in agreement with the majority
position in the literature in that German has an overt VP (vP )-internal subject
position. His claim diverges from the mainstream, however, in that he denies the
existence of a VP-external subject position in the Mittelfeld, i.e. SpecTP. Haider
directly criticizes the superimposition of English phrase structure rules onto Ger-
man and argues against obligatory movement of the subject to SpecTP. We will
side with him and the majority of the literature on this point. However, we will
not agree with his contention that German essentially has a flat Mittelfeld with no
TP-projection at all. He posits that the assumption that there is only a VP-internal
subject position implies the following corollary:
Es gibt im Deutschen, ceteris paribus, keine durch die strukturelle Posi-
tion bedingte Subjekt-Objekt-Asymmetrie hinsichtilich der Extraktions-
dom?anen zwischen Subjekten und Objekten. (Ceteris paribus, German
does not show structurally determined subject-object asymmetries with
regards to the extraction domain between subjects and objects, my trans-
lation JJ) (Haider 1993, p.150)
He then goes on to argue that this corollary is indeed borne out. However, both
the logic and the empirical accuracy of this reasoning can be questioned. First, the
assumption that being located within the same maximal projection solely determines
whether subextraction is possible or not is highly theory-dependent. Even under a
Barriers-style theory this would only be true if Government was defined in terms of
m-command. V would then both govern its complement and its Spec. A definition
of Government in terms of c-command would not give this result, as the specifier
of VP is clearly not c-commanded by V. The same reasoning can be applied to
44
L-marking. If L-marking is tied to theta-marking (defined in terms of m-command)
the argument goes through, otherwise it doesn?t. M-command might ultimately
turn out to be a relevant relation in the grammar but there is no a priori reason
why we should assume it, yet it is crucial for Haider?s corollary to go through.
Second, and this brings us directly back to the main theme of this dissertation,
the claim that extraction out of VP-internal subjects and objects is equally accept-
able does not hold up to close empirical scrutiny. We will return to the relevant
examples and the treatment they received later in this section and in 3.2.6
Concluding our discussion of the subject in the German Mittelfeld, we assume
for concreteness, and paralleling a standard analysis of the English clause that this
derived position is SpecTP, but it could just as well be identified as SpecIP, SpecA-
grP etc. Nothing hinges on this notational difference. What is crucial is that this
position is a derived position external to VP (and vP ), which a subject has overtly
moved to. I will refer to this position as ?derived subject position?.
6The only analysis of German that denies that indefinite subjects can remain VP (vP)-internal
that I am aware of is due to Bobaljik & Jonas (1996). Extending their discussion of subjects in
Icelandic, they claim that the subject positions in German are SpecAgrsP and SpecTP, i.e. both
are outside VP and subjects in German can never remain in-situ. Their argument is based on the
observation that the subject cannot possibly follow manner adverbs.
i. Es
there
essen
eat
Kinder
children
sorgf?altig
carefully
?Apfel.
apples
?There are children carefully eating apples.?
ii. *Es
there
essen
eat
sorgf?altig
carefully
Kinder
children
?Apfel.
apples
?There are children carefully eating apples.?
While Bobaljik & Jonas?s (BJ) observation is intriguing there are a number of reasons why I am
not particularly worried about it. First, BJ themselves admit that ?various complications of course
arise? but that ?the positions delineated by Diesing are at the very least amenable to the analysis
we have proposed for Icelandic [...] since this is the claim the theory requires, it is the one we will
adopt.? Second, Pure Manner Adverbs (Ernst 1984), such as sorgf?altig, could plausibly be argued
to be modifiers of VP rather than vP . The adverb would then be expected to follow even in-situ
external subjects. While BJ?s discussion is interesting, we take it to be a minority position and
will follow the vast majority of the literature in assuming that subjects can in fact remain in-situ
in German.
45
3.1.2.3 Internal and external subjects
Additionally, I am also distinguishing between internal and external subjects. This
dichotomy goes back to Burzio (1981) and Williams (1981), who introduce the
distinction between external and internal theta roles. Internal subjects standardly
receive theme theta roles, and are the sole arguments of unaccusative and passive
predicates. They are located in the complement position of V. External subjects
are the subjects of unergative predicates and are taken to receive agent theta roles.
What unifies them is that they can both be the only argument in a clause, they
standardly receive nominative case (in nominative-accusative languages).
There is some controversy as to their exact structural position. Up to the mid
eighties no distinction was made between external subjects and derived subjects.
The standard subject position was the specifier of S. Early GB saw the advent of the
VP-internal subject hypothesis (early references include Koopman& Sportiche 1985,
1991 and Kitagawa 1986, see McCloskey 1997:203ff. for a more detailed chronology).
The external subject was now assumed to originate inside the VP in a position either
identified as SpecVP or adjoined to VP (as in Koopman & Sportiche?s original
proposal).
Under this early version of the VP-internal subject hypothesis, subjects and
objects were taken to bedominatedby thesame maximal projection. Kratzer (1996),
building on observations made by Marantz (1984), encodes the intuition that objects
have a closer relationship with the verb than subjects and proposes to base-generate
subjects in the Spec of a voice-projection, which takes VP as its complement. A
similar approach was taken by Chomsky (1995), who generates external subjects in
the Spec of vP . Following recent analyses of German phrase structure (Grewendorf
2002, M?uller 2004, 2010) we identify the base position of the external subject as
SpecvP . The reader should note, however, that Kratzer?s SpecVoiceP or SpecVP7
7It is less clear whether the external subject position could be associated with an adjunct
46
would be possible alternatives and compatible with the claims made throughout this
thesis.
In this section we have summarized some of the standard assumptions made
in the literature which lead us to the conclusion that German has four positions
where subjects can appear in overt syntax: (i) a topic position in SpecCP, (ii) a
derived position in SpecTP, (iii) an external position in SpecvP and (iv) an internal
position as the complement of V. This is illustrated in the tree diagram in (5):
(5) CP
topic TP
derived subject vP
external subject VP
internal subject V
3.1.3 A brief history of German NP-subextraction
This section provides a brief overview of subextraction in German, focusing on the
was-f?ur split construction. Working through the literature on this topic can be
somewhat dispiriting. While there is often great theoretical insight, there is also a
remarkable lack of agreement on some of the key judgments. We will first discuss
the data with the (sometimes contradictory) judgments as reported by the authors,
setting the stage for more careful experimental investigations of the facts in the later
sections of the chapter.
position to VP, as Koopman & Sportiche?s original proposal assumed. If that were the case we
might expect, ceteris paribus, that extraction out of external subjects should be as degraded as
extraction out of adjuncts. Our studies laid out in detail below rather suggest that subextraction
out of external subjects has an intermediate status and is not as degraded as extraction out of
adjuncts.
47
The what for construction is found in a number of Germanic languages in-
cluding German, Dutch, Danish, Swedish, Norwegian and Yiddish. It is normally
glossed as what kind of NP. What is relevant for our purposes here is that the con-
struction has the property of allowing a split of the what and the for NP part, as
illustrated in (6).
(6) a. Was f?ur ein Schnitzel
what for a schnitzel
hat
has
der
the
Hermes
Hermes
verspeist?
eaten.up
[was-f?ur ]
?What kind of schnitzel did Hermes eat up??
b. Was
what
hat
has
der
the
Hermes
Hermes
f?ur ein Schnitzel
for a schnitzel
verspeist?
eaten.up
[was-f?ur split ]
?What kind of schnitzel did Hermes eat up??
While having received a considerable amount of attention on its own account (see
Leu 2008 for an overview and references), was-f?ur split is often used as a diagnostic
for the islandhood of certain domains. While the details of the internal structure of
the was-f?ur phrase differ, virtually all analyses assume that genuine subextraction
of the ?was? element has occurred. We will follow this majority position and use was-
f?ur split as our main diagnostic for testing the permeability of specific structural
positions in German. was-f?ur split has the advantage vis `a vis subextraction of PP-
complements out of NPs that the latter do not unambiguously indicate that actual
extraction has taken place since German does not allow stranding the preposition.
We will see in our discussion of PP subextraction in English in section 5.2 that
pied-piping has a significant effect on the acceptability of subextraction, and we will
conclude that only in the case of P-stranding did genuine extraction occur, while
pied-piped PPs do not originate within the NPs. With was-f?ur split this problem
does not arise.
48
3.1.3.1 den Besten 1985
The discussion of was-f?ur split in German is virtually synonymous with the dis-
cussion of NP-subextraction in general, as was-f?ur split is the construction used to
investigate subextraction domains. The first detailed investigation of in the gen-
erative literature goes back to den Besten (1985). He bases his discussion on the
following data.
(7) a. *Was
what
haben
have
f?ur
for
Leute
people
deine
your
Mutter
mother
besucht?
visited
[subject]
?What kind of people visited your mother??
b. *Was
what
hast
have
du
you
f?ur
for
Leuten
people
deinen
your
Aufsatz
paper
geschickt?
sent
[indirect object]
?What kind of people have you sent your paper to??
c. Was
what
hast
have
du
you
in
in
Italien
Italy
f?ur
for
Museen
muesums
besucht?
visited
[object]
?What sort of museums did you visit in Italy??
d. Was
what
sind
are
gestern
yesterday
f?ur
for
merkw?urdige
weird
Sachen
things
passiert?
happened
[unaccusative]
?What kind of weird things happened yesterday??
He argues that the gap resulting from was-f?ur split is only licensed when the was-f?ur
phrase is in the internal argument position, i.e. a complement of V. As a result,
was-f?ur split is allowed out of objects (7-c) and internal subjects, that is subjects
of unaccusatives, (7-d), and passives, but not out of subjects of unergatives, (7-a),
or out of indirect objects, (7-b). This is illustrated in (8) (den Besten?s (39)):
49
(8) ?S
COMP
was
S
NP
[t f?ur ...]
VP
NP
[t f?ur ...]
?V
NP
[t f?ur ...]
V
BW
?
?
How do den Besten?s argument positions relate to the phrase structure given in (5)?
He explicitly advocates for a VP-internal subject position. This position, however,
is not identical to Koopman and Sportiche?s VP-internal subject hypothesis but
rather identifies the position of the arguments of unaccusatives and passives, i.e.
our internal subject position (the complement of V). External arguments are not
base-generated in VP but in SpecS, i.e. the notions external subject and derived
subject are conflated. What appears to be SpecVP in (8) is in fact the position of
the indirect object.
At the end of this chapter, after having gone through a series of experiments,
we will conclude that den Besten?s generalization is right on the mark. Essentially,
we find a complement vs. everything else dichotomy. But first things first. We
have to deal with the undeniable fact that den Besten?s judgment?s did not stand
uncontested.
To my knowledge, no one questioned the acceptability of was-f?ur split out of
objects, but examples of both subextraction out of external subjects and indirect
object are sometimes reported to be acceptable:
50
(9) a. Was
what
hast
have
du
you
f?ur
for
einem
a.DAT
Buch
book
einen
an
Aufsatz
article
zugef?ugt?
added
?What (kind of) book did you add an article to??
[Bayer et al. 2001, ex. 42]
b. Was
what
hat
has
sie
she
f?ur
for
Leuten
people.DAT
ein
a
Buch
book
gegeben?
given
?What kind of people did she give a book to??
[Lutz 2001, ex. 148b]
c. Was
what
hat
has
ihn
him
f?ur
for
ein
a.NOM
Chefredakteur
chief.editor
angerufen?
called
?What kind of chief editor called him??
[Lutz 2001, ex. 148a]
(9-a) and (9-b) directly contradict (7), where virtually identical examples are re-
ported to be unacceptable. So who?s right? Or are both right and are we just
looking at two different idiolects? The short answer is that this is a domain where
informal introspective gathering of data will not suffice. The reader is referred to
general problems with the informal acceptability judgments in section 2.2.
3.1.3.2 Melting
(9-c) is in conflict with (7-a), but differs from it in one interesting and very relevant
respect. In the former the direct object has been scrambled across the subject
that was extracted out of. This phenomenon has been dubbed Melting by M?uller
(2010), who reports that scrambling of an element unfreezes a certain domain (earlier
discussion of this phenomenon goes back to Diesing 1990 and de Hoop 1996). M?uller
illustrates the contrast with the following examples:
51
(10) a. *Was
what
haben
have
denn
PRT
f?ur
for
Leute
people
den
the
Fritz
Fritz
getroffen?
met
?What kind of people met Fritz??
b. Was
what
haben
have
den
the
Fritzi
Fritz
denn
PRT
f?ur
for
Leute
people
ti getroffen?
met
?What kind of people met Fritz??
[M?uller 2010, p.61]
Extraction out of external subjects is generally disallowed, as indicated in (10-a),
but the subextraction can be salvaged by scrambling the direct object across the
subject, as shown in (10-b). M?uller aims at deriving CED effects from the Phase
Impenetrability Condition (PIC) (Chomsky 2000). Given certain assumptions on
the mechanics of feature checking and phases8, any XP that is the last element
merged in a given phase should be an island for extraction. In (10-b), however, the
object is moved to the edge of the vP phase, which effectively pushes the subject
down one spot re-ranking it as the penultimate element of the phase. As a result,
the subject ceases to be an island.
M?uller ?s theoretical account and his prediction of Melting effects is intriguing,
but it is not clear whether his effects hold as reported. We will return to the
discussion of these effects and we will scrutinize the empirical accuracy of Melting
in an experiment in section 3.5.
3.1.3.3 Diesing 1992
M?uller ?s judgments, as he himself mentions in a footnote (p.68, fn.43), diverge from
what Diesing (1990, 1992) reports. She lays emphasis on the importance of the
subject?s position for purposes of extractability by observing that splits from in-situ
8Specifically, M?uller stipulates that features on a head are ranked with respect to each other,
that edge features triggering successive cyclic movement must be added before the phase head
becomes inert, and that movement proceeds through the edge of every XP. The reader is referred
to the paper for a detailed discussion about how these specific technical assumptions yield the
desired results.
52
subjects are preferred over splits from moved subjects, as illustrated in (11) (Diesing
1990, p. 55 attributes this observation to Angelika Kratzer in class lectures of 1988):
(11) a. Was
What
haben
have
denn
PRT
f?ur
for
Ameisen
ants
einen
a
Postbeamten
postman
gebissen?
bitten
?What kind of ants bit the postman??
b. *Was
What
haben
have
f?ur
for
Ameisen
ants
denn
PRT
einen
a
Postbeamten
postman
gebissen?
bitten
?What kind of ants bit the postman??
Note that (11-a) and (10-a) are virtually identical examples, modulo the choice of
lexical items. Yet they are given opposite judgments. Again we may ask: Who?s
right? Which judgment should we base our theory on? The answer again is that
this a domain where our informal methods of gathering data crumble.
Returning to Diesing, we remind ourselves that anindefinite subject in German
can either move to SpecTP or stay in situ in SpecvP, if the SpecCP position is filled
by another element. The particle denn is used to demarcate the TP from the VP
domain to determine the position of the subject. In (11-a) the subject remains in-
situ and the split is judged acceptable, whereas in (11-b) the subject is moved and
the example is unacceptable. Diesing concludes that Huang?s CED only applies to
subjects in SpecTP (derived subjects) while subjects in VP are not subject to the
CED.
She provides an analysis in terms of a modified version of Chomsky?s (1986)
Barriers system. Concretely, Diesing stipulates that aspectual verbs like have (or
German haben) ?-mark and hence L-mark their complements. Furthermore, she
makes the assumption that ?if a head L-marks a maximal projection, it L-marks
the specifier of that projection? (Diesing 1990, p.52).9 As a result, both VP and the
9A similar mechanism is considered by Chomsky and Koopman&Sportiche (1991) for case
assignment with ECM predicates. This predicts, however, that the accusative marked thematic
subject of the lower clause should be transparent for subextraction. Diesing (1990, p.86) claims
that there is a small contrast between [i] and [ii], while conceding that [i] is still degraded.
53
subject in SpecVP are exempt from barrierhood and extraction is possible. TP, on
the other hand, is not L-marked, and as a result the subject in SpecTP cannot inherit
the L-marking status. Consequently, both TP and the Spec of TP are barriers (by
virtue of being blocking categories) and extraction out of the subject in SpecTP is
blocked.
There are number of comments and objections that need to be made at this
point. First, Diesing?s analysis inherits all the well known problems associated
with the Barriers system (see Lasnik&Saito 1992 for discussion). Second, she takes
advantage of the rather vague definition of ?-marking and extends it to what she calls
?aspectual verbs?. If that were the case, we would expect this pattern of acceptability
to be directly tied to the presence of an auxiliary verb. This is however not the case
in the was-f?ur split cases Diesing discusses:
(12) a. Was
What
bei?en
bite
denn
PRT
f?ur
for
Ameisen
ants
einen
a
Postbeamten?
postman
?What kind of ants bit the postman??
b. *Was
What
bei?en
bite
f?ur
for
Ameisen
ants
denn
PRT
einen
a
Postbeamten?
postman
?What kind of ants bit the postman??
The examples in (12) have the same status as Diesing?s in (11), as a controlled
acceptability judgment study confirms10. To get the contrast in (12) we would have
to modify Diesing?s proposal to allow for the VP to be L-marked by T across the
board. It is left to the reader to decide how natural of an assumption that would be
and how other islands phenomena could be covered if VP was generally exempted
i. ??Who do you believe pictures of to be on sale?
ii. *Who are pictures of on sale?
Chomsky (2008) argues that [i] is in fact acceptable, diverging from Chomsky (1973) where both
(i) and (ii) were deemed unacceptable. We will come back to this issue in detail in section 5.3 and
conclude from an empirical study that both [i] and [ii] are equally degraded and that this is fully
expected under a general theory of freezing.
10Data analysis in progress.
54
from barrierhood.
Ironically however, there is a set of examples not discussed by Diesing where
the islandhood of a subject seems to exactly rely on the presence of an auxiliary.
Extraction out of non-finite sentential subjects in German, as noted by Grewendorf
(1989), improves when the V2 position is filled with an auxiliary as compared to a
main verb. I will only mention these facts here and return to a detailed empirical
and theoretical discussion in section 4.4.
(13) a. Wessen
whose
Beispiele
examples
hat
has
zu
to
analysieren
analyze
dich
you
mehr
more
frustriert
frustrated
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
?(*?) Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
b. *Wessen
whose
Beispiele
examples
frustrierte
frustrated
dich
you
zu
to
analysieren
analyze
mehr
more
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
(*?)?Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
[Grewendorf 1986, p. 66-67]
Returning to Diesing?s account of the cases in (11), I have to conclude that her
analysis is unsatisfactory. Kratzer?s observation of the contrast between external
and derived subjects is intriguing, and it is exactly the sort of contrast that led
Stepanov (2007) to conclude that CED effects can be reduced to freezing effects.
This is a reasonable conclusion based on the data he had at his disposal. I will show,
however, in what follows that both den Besten and Diesing only present part of the
facts and that a closer scrutiny of the data requires us to maintain a theory that
accounts for complement/non-complement asymmetries as well as an explanation of
freezing effects.
55
Table 3.1: Which domains allow subextraction in German?
SpecTP Indirect Object SpecvP internal argument
den Besten (1985) N/A ? ? BW
Diesing (1992) ? N/A BW BW
Haider (1993) N/A BW BW BW
Lutz (2001) N/A BW BW BW
M?uller (2010) ? ?1 ?1 BW
1 Only judged acceptable if another constituent has been scrambled across.
3.1.3.4 Conclusion
What have we learned from the discussion in this section? Clearly, there has been
much insightful work on subextraction in German. Yet, the status of some of the
crucial facts we are building our theories on remains unclear. The range of disagree-
ments discussed in the last section are summarized in Table 3.1.
We see that there is conformity on the extreme sides of the scale: subextrac-
tion out of internal arguments is judged uniformly good while subextraction out
of derived subjects is judged uniformly bad (by those authors that considered this
case). Unfortunately, the clear cases also are the least theoretically controversial.
Every theory I am aware of, both in terms of freezing and a theory in terms of
CED-effects, makes the same predictions regarding these cases. The crucial exam-
ples involve subextraction out of external subjects and indirect objects.
3.2 Experiment 1 - External vs. derived subjects
3.2.1 Introduction
In our review of the theoretical work on constraints on subextraction in German
we found the status of the empirical facts to be dissatisfying. The purpose of the
56
Figure 3.1: Predictions for subextraction out of subjects/objects: left ?pure freezing?
predictions; right freezing and CED effects
gid12gid8gid7gid9gid13gid5gid4gid1
gid9gid2gid6gid5gid3gid11gid1
gid7gid9gid13gid5gid4gid1
gid9gid2gid6gid5gid3gid11gid1
gid12gid8gid7gid9gid13gid5gid4gid1
gid10gid12gid2gid6gid5gid3gid11gid1
gid7gid9gid13gid5gid4gid1
gid10gid12gid2gid6gid5gid3gid11gid1
gid12gid8gid7gid9gid13gid5gid4gid1
gid9gid2gid6gid5gid3gid11gid1
gid7gid9gid13gid5gid4gid1
gid9gid2gid6gid5gid3gid11gid1
gid12gid8gid7gid9gid13gid5gid4gid1
gid10gid12gid2gid6gid5gid3gid11gid1
gid7gid9gid13gid5gid4gid1
gid10gid12gid2gid6gid5gid3gid11gid1
following experiments is to collect more fine-grained data that will turn out to be
decisive in the ongoing theoretical debates. We will start with the contrast between
subextraction out of external subjects, derived subjects and out of objects, using
was-f?ur split .
3.2.2 Predictions
We saw above that Stepanov (2007) contends that subject island effects are reducible
to freezing effects. If this is thecase we would expect the data to show a split between
extraction out of moved domains vs. extraction out of unmoved domains. In other
words, we would not expect to find a main effect based on whether the extraction
originated from a subject or from an object. However, that is exactly the prediction
the CED would make. If, on top of that, Diesing?s observation is correct in that
there is an asymmetry between extraction out of in-situ vs. moved subjects, we
would expect a three way distinction: extraction out of objects > extraction out of
unmoved subjects > extraction out of moved subjects. These different predictions
are schematized in Figure 3.1. The chart on the left shows the results that would
be predicted by a freezing-only theory, i.e. if there is only a freezing effect but no
subject/object asymmetry, and the chart on the right shows results that would be
expected if there are two separate main effects.
57
3.2.3 Methodology
3.2.3.1 Participants
32 self reported native speakers of German11 participated in the experiment. The
participation was unpaid and voluntary. 28 undergraduates at the University of
Vienna with no training in formal syntax were asked to rate sentences presented
to them in a paper questionnaire. 4 speakers did the experiment online. Alex
Drummond kindly allowed me to use his webspr software (http://code.google.com/
p/webspr/) and helped me with the set up code.12 1 speaker was filtered for not
completing the questionnaire. The data of 31 participants was used for data analysis.
3.2.3.2 Procedure
Participants were aware that they were taking part in a linguistics experiment,
but were left uninformed about what was being tested. They were asked to rate
sentences on a 7-point Likert scale according to their native speaker intuitions. The
scale was anchored, i.e. participants were asked to give a 6 or a 7 to sentences they
found perfectly acceptable, to give a 1 or a 2 to sentences they found completely
unacceptable and to give 3 -5 to sentences they found not totally unacceptable but
also not completely perfect. The reader is referred to the discussion in section 2.2.4
for reasons why a Likert scale was chosen over the Magnitude Estimation technique.
In the instructions for the experiments, it was emphasized that prescriptive
rules and plausibility of the sentences were irrelevant for the experiment. Speakers
11The large majority were speakers of Standard Austrian German. There was no motivation
for this choice other than the fact that those speakers were most readily accessible. While it is
possible that was-f?ur split is not available to the same degree in every dialect of German, there is
no reason to believe that the relative acceptability across the various conditions should differ.
12In all experiments except for this one his online software was used exclusively. This speeds up
the data collection process significantly as compared to paper questionnaires and eliminates one
potential sourceof human error, namelythe investigatormaking mistakescopying the questionnaire
results to an Excel sheet. Also, check out Alex Drummond?s new and even more user-friendly
interface Ibexfarm under http://spellout.net/ibexfarm.
58
were also given example sentences: one perfectly acceptable sentence (a regular wh-
question) judged with a 7, one completely unacceptable sentence (a CSC violation)
judged with a 1, and one intermediate sentence (a wh-island violation) judged with
a 3. Participants also signed a statement indicating that they were over 18 years
of age. They accepted that this experiment was not intended for their benefit but
solely for research purposes, and it was made clear that their data was analyzed
anonymously. See the Appendix for German, English, Japanese and Serbian versions
of the instructions.13
3.2.3.3 Design
The experiment tested was-f?ur split , manipulating thefactorsSub/ObjandMoved/
InSitu. 2 baseline conditions where not split takes place were added. This gives us
the following 6 conditions.
(14) a. Was
What
f?ur
for
eine
a
Ameise
ant
hat
has
denn
PRT
den
the
Beamten
clerk
gebissen?
bitten
[sub, -split]
?What kind of ant bit the clerk??
b. Was
What
f?ur
for
einen
a
Beamten
clear
hat
has
denn
PRT
die
the
Ameise
ant
gebissen?
bitten
[obj, -split]
?What kind of clerk did the ant bite??
c. Was
What
hat
has
denn
PRT
f?ur
for
eine
an
Ameise
ant
den
the
Beamten
clerk
gebissen?
bitten
[in-situ sub, +spl]
?What kind of ant bit the clerk??
d. Was
What
hat
has
f?ur
for
eine
an
Ameise
ant
denn
PRT
den
the
Beamten
clerk
gebissen?
bitten
[mvd sub, +spl]
?What kind of ant bit the clerk??
e. Was
What
hat
has
denn
PRT
die
the
Ameise
ant
f?ur
for
einen
a
Beamten
clerk
gebissen?
bitten
[in-situ obj, +spl]
?What kind of clerk did the ant bite??
f. Was
What
hat
has
denn
PRT
f?ur
for
einen
a
Beamten
clerk
die
the
Ameise
ant
gebissen?
bitten
[mvd obj, +spl]
?What kind of clerk did the ant bite??
13Feel free to use for your own studies but please reference.
59
In (14-a) and (14-b) the entire was-f?ur phrase is moved to the left periphery and no
split takes place. In (14-c) was is extracted out of the in-situ subject, whereas in
(14-d) it is extracted out of a moved subject position. Following Webelhuth (1989)
and Diesing (1990, 1992), the particle denn (?indeed?) is used to detect whether the
subject in its base or in a derived position (see sections 3.1.2 and 3.1.3 above for
further discussion). In (14-e) the split originates from an unmoved object and in
(14-f) was is moved out of an object in a derived position.
We created 3 lexicalizations. for each condition yielding 108 items total, which
were distributed among 6 lists in a Latin Square design. This method makes sure
that potential noise caused by specific lexicalizations is distributed across condi-
tions and does not affect one condition disproportionately. The resulting 18 stimuli,
together with 24 stimuli from a different experiment and 36 fillers of all levels of un-
acceptability14 were pseudo-randomized. As a result each participant was presented
with 78 sentences.
All predicates used were transitives. The complete list of items can be found
in the appendix. It must be noted that animacy of the subject and object were not
controlled for in the original experiment. A follow-up experiment was rerun to make
sure that this is not a confounding factor. The follow-up experiment confirmed the
results: All effects persist if animacy is controlled for. This is related to a point
raised by Valentine Hacquard (p.c.), which is relevant for all acceptability judgment
studies using was-f?ur split . There is a sense in which it is not equally felicitious to
ask a was-f?ur question with any type of NP or in any kind of context. Hacquard
points out that it might be stranger to ask about the type of clerk in ant biting
incident than the type of clerk. I very much share this concern, which makes it
all the more crucial to compare to always have control conditions where no was-f?ur
14We used finite sentential arguments varying the factors Sub/Obj and Extraction, Coordi-
nate Structure Constraint (CSC) violations, complex NP islands, extraction across verba dicendi
as well as regular wh-questions. See the appendix for a full list of fillers.
60
split takes place. Whatever the pragmatic oddity of inquiring the kind of a given
NP, it will also show up in the control condition. There is no reason to believe
that the split should increase this pragmatic effect. As we are only looking at the
interaction effects, i.e. the relative decrease caused by was-f?ur split , this pragmatic
concern should not be an interfering factor.
3.2.4 Results
Our results show a pattern as expected under the CED. They do not pattern accord-
ing to Stepanov?s predictions. While Diesing?s (1992) observation is confirmed that
was-f?ur split out of in-situ subjects is more acceptable than out of moved subjects
(3.55 vs. 2.28, t(1,92)=5.2, p<.001), our data reveal another interesting pattern:
extraction out of the in-situ subject is significantly degraded compared to in-situ
objects (t(1,92)=11.2, p<.001), and extraction out of moved subjects is marginally
degraded compared to unmoved subjects (t(1,92)=2.4, p=.09). Informally speak-
ing, acceptability decreases when you extract out of a subject or when you extract
from a moved domain. The effect is cumulative, i.e. extraction out of a moved
subject leads to the worst results. This is summarized in Figure 3.2, the descriptive
statistics in Table 3.2 and the ANOVA15 data results in Table 3.316 .
3.2.4.1 Distribution of the data
A common criticism brought up against acceptability judgment studies of this sort
(e.g. Den Dikken et al. 2007) is that averaging the data of different speakers might
obscure the fact that people have different grammars, i.e. different I-languages (cf.
15It is sometimes pointed out that Likert scales yield non-parametric ordinal data, which is
strictly speaking not amenable to ANOVAs, as ANOVAs assume normally distributed interval
level data. However, it is a standard practise in much of the psychology literature to analyze this
kind of data using ANOVAs, especially since no non-parametric tool of equivalent statistical power
is available. We will follow this convention, acknowledging this caveat. See Sprouse (2007) for
further discussion.
16For all ANOVA results a notational convention is used: ??? = p < .001, ?? = p < .01
61
gid13gid1
gid14gid1
gid15gid1
gid16gid1
gid17gid1
gid18gid1
gid19gid1
gid3gid10gid1gid11gid8gid7gid5gid1 gid2gid10gid1gid11gid8gid7gid5gid1 gid3gid10gid1gid20gid8gid7gid5gid10gid1gid11gid6gid9gid4gid1 gid3gid10gid1gid20gid8gid7gid5gid10gid1gid20gid6gid9gid4gid1 gid2gid10gid1gid20gid8gid7gid5gid10gid1gid11gid6gid9gid4gid1 gid2gid10gid1gid20gid8gid7gid5gid10gid1gid20gid6gid9gid4gid1
Figure 3.2: Experiment 1: Results
the discussion in section 2.2.1). While this is a prima facie possibility, it is not what
a closer look at the data suggests.
The graph in Figure 3.3 presents the individual results for the relevant condi-
tions of all 31 participants of the was-f?ur split experiment. The black column to the
left is the split out of the unmoved object condition, (14-e), the dark grey column in
the middle presents the judgments for the subject, in-situ condition, (14-c), and the
light grey column to the right is the subject, moved condition, (14-d). It is clearly
visible that the overall pattern of acceptability is constant for a large majority of
the participants. The judgments of 28 out of 31 (= 90.3%) speakers directly reflect
the average pattern across speakers for the relevant conditions, i.e. extraction out
of objects is preferred to extraction out of unmoved subjects, which is preferred to
extraction out of moved subjects. For 3 speakers, extraction out of moved subjects
was rated more acceptable than extraction out of unmoved subjects17 and every
17It seems likely that these judgments are noise as a result of the fact that only three lexicaliza-
62
Table 3.2: Descriptive statistics
Mean SD
Sub, -split 6.35 1.21
Obj, -split 6.64 0.83
Sub, +insitu 3.55 1.81
Sub, -insitu 2.28 1.51
Obj, +insitu 6.14 1.29
Obj, -insitu 2.84 1.71
Table 3.3: ANOVA
F p
Moved/InSitu 222.023 ***
Sub/Obj 113.554 ***
Sub/Obj x InSitu 44.906 ***
speaker gave extraction out of objects the highest rating.
These results quite clearly go against the possibility that we are producing
a meaningless average across different idiolects. The overall pattern is reflected in
almost all individuals, which suggests that they all share the same grammatical con-
straints. The fact that the absolute numbers vary across speakers is not surprising
at all. There is a considerable amount of inter-speaker variation, i.e. noise, which
is due to a number of factors including the position of the stimulus within the ex-
periment (beginning, middle or end and the adjacent items), which lexicalization
of which condition18 a speaker is presented with and the individual?s concentration
span. In a different experiment multiple tokens of the same item were used and it
was not uncommon that speakers gave varying judgments to identical stimuli. All of
this points to the conclusion that the differences in absolute judgments we observe
do not reflect differences in individual grammars, but are noise inherent to the task
of making judgments.
tions were given for each condition. Using 6 or even 9 lexicalizations is likely to further reduce the
noise so that 100% of the speakers behave according to the overall pattern. However, given that
the overwhelming majority of participants shows uniform behavior, this issue seems negligible.
18While every attempt was made to make all items similar in length and pragmatic plausibility,
there is no guarantee that some lexicalizations of the same condition sound better to a speaker
than others. The Latin square design makes sure that this does not disproportionally affect a
single condition but only causes noise equally distributed across the conditions.
63
Figure 3.3: Experiment 1: Individual results of all 31 participants
64
3.2.5 Discussion
Our results areunsurprising inthat theyshow that bothsubject andobject questions
without subextraction are rated as highly acceptable across the board. In addition,
it comes as no surprise that we see some drop when the was is subextracted. Fronting
the was on its own creates a filler-gap dependency, which is well known to yield an
overall decrease in acceptability. The parser needs to store the filler in memory until
it sees a gap site where it can be discharged. This increased processing load leads
to a decrease in o?ine acceptability (cf. Gibson 1998).
The three-way distinction in extractability we find has gone undetected so far
in the theoretical literature. It strongly suggests that not one, but two constraints
are active in the grammar of German, which have to be held separately.
(15) Constraints active in German
a. Extraction out of moved domains is degraded (= Freezing Effect)
b. Extraction out of subjects is degraded (=Subject Condition)
What this experiment shows is that, crucially, (15-b) cannot be reduced to (15-a),
contrary to Stepanov?s claim. Moreover, the constraints are cumulative in the sense
that violating both constraints, i.e. extracting out of moved subjects leads to the
lowest acceptability. Extracting out of moved objects, violating (15-a) but not
(15-b), receives the second lowest rating. Extracting out of unmoved subjects, vio-
lating (15-b) but not (15-a), is judged around 3.55, and extraction out of unmoved
objects, violating neither (15-a) nor (15-b), receives the highest rating with 6.14.
We will show in section 3.6.2 how the subject/object asymmetry can be ac-
counted for by Uriagereka?s (1999) MSO theory. As we saw in section 2.1.2.1, it
would also account for the decreased acceptability of extracting out of the moved
object, (14-f), as the movement operation would essentially trigger the object to be
65
spelled-out before subextraction can take place. MSO has nothing to say about the
difference between extraction out of moved vs. extraction out of in-situ subjects.
There is no shame, however, in complementing it with a separate account of freez-
ing effects, e.g. Takahashi (1994) or Hunter (2010). It seems plausible that the two
separate effects we find have two independent sources.
3.2.5.1 Excluding extra-grammatical factors
The case made in section 2.2.4 was that parts of the gradience observed in accept-
ability judgments has its source in the grammar. How do we know that the effects
observed in this experiment are of this kind? Let us take a closer look at the two
key examples, was-f?ur split from an in-situ subject in (14-c) and object in (14-e)
(repeated here), and what differences we can detect other than the complement/non-
complement asymmetry of the extraction site.
(16) a. Was
What
hat
has
denn
PRT
f?ur
for
eine
an
Ameise
ant
den
the
Beamten
clerk
gebissen?
bitten
?What kind of ant bit the clerk??
b. Was
What
hat
has
denn
PRT
die
the
Ameise
ant
f?ur
for
einen
a
Beamten
clerk
gebissen?
bitten
?What kind of clerk did the ant bite??
(16-a) and (16-b) are exactly the same length, in fact they are composed of identical
lexical items. The distance between the wh-filler and a gap is known to affect
acceptability. Longer distance typically decreases the acceptability, as the parser
needs to store the filler in memory longer (cf. Gibson 1998). Looking at the filler-
gap distances in the examples at hand, we find exactly the opposite picture. No
matter which metric of distance is employed (number of words, number of XPs,
depth of embedding) the dependency in (16-b) is longer than in (16-a). Whatever
the cost of dependency formation is, it is clearly outweighed by the grammatical
66
difference between subjects and objects.
Another factor that also goes in the opposite direction of the effects we find
is a brief local ambiguity at was, which at this point could be a subject or an
object wh-element, a wh-scope marker or part of was-f?ur split .19 This ambiguity is
consistent across the conditions. The disambiguation and potential reanalysis has
to take place at f?ur which forces the was-f?ur split parse, as f?ur-PPs cannot appear
in argument positions. If anything, this might cause a more significant problem in
the object case as the disambiguating element appears later in the string. Whatever
this cost is, it is again outweighed by the grammatical distinction. We conclude
that both the distance of the filler-gap dependency and reanalysis, two well-known
sources of decreased acceptability stemming from the parser, cannot be responsible
for the acceptability pattern we see. Whatever cost they incur is easily outweighed
by the grammatical differences.
Let us now turn to two potentially interfering factors that might skew the
data in favor of object subextraction. One domain which is sometimes claimed
to affect the acceptability of extraction is information structure. Let us look at
one representative proposal. Goldberg (2006), pursuing the program of reanalysing
island effects in purely extra-grammatical terms, offers the following explanation:
Elements in unbounded dependencies are positioned in discourse promi-
nent slots. It is pragmatically anomalous to treat an element at once
backgrounded and discourse prominent.
[Goldberg 2006, p. 135, original italics]
In other words, Goldberg suggests that A?-dependencies cannot originate in topic
or old information domains. Lidz & Williams (2009, 184) point to two very obvious
problems with this proposal. They note that relative clauses show the same island
sensitivity as wh-clauses but no discourse prominence is associated with the relative
clause head. This argument can easily be illustrated with a subject island violation.
19In certain dialects of Southern German was could also be an adjunct question meaning ?why?.
67
(17) a. I know the politician that John wrote a book about.
b. *I know the politician that a book about caused a scandal.
Second, Lidz & Williams note that no pragmatic anomaly is associated with putting
focused elements in presuppositional, i.e. backgrounded, contexts, as their example
(12) shows.
(18) I certainly did not read the book that CHOMSKY recommended.
Third, we can add to this the well known fact that wh-insitu languages allow wh-
elements inside islands. The following example shows a wh-element in Japanese,
which can appear inside a relative clause but cannot be scrambled out of it.20
(19) a. Mary-wa
Mary-top
[John-ga
[John-nom
dare-ni
who-Dat
ageta
gave
hon]-o
book]-Acc
hirotta-no?
picked.up-Q
?(*) Mary pick up the book that John gave to who??
b. *Dare-ni
who-Dat
mary-wa
Mary-top
[john-ga
[John-nom
t ageta
gave
hon]-o
book]-Acc
hirotta-no?
picked.up-Q
?(*)Who did Mary pick up the book that John gave to??
We conclude that Goldberg?s analysis considerably undergenerates and cannot be
maintained in its current form. For the sake of argument, however, let us assume
that it is indeed true that for some poorly understood reasons constituents conveying
new information allow extraction the most easily. Let us assume furthermore that
elements occurring later in the string are canonically more likely to be interpreted
as focus, whereas earlier constituents are canonically interpreted as topics. If all
of these assumptions hold, this could be a potentially confounding factor and be
(part of) the reason why speakers disprefer extraction out of subjects compared to
objects.
Finally, (16-a) and (16-b) also differ with respect to the distance between the
20Thanks to Maki Kishida for this example.
68
gap and the verb. The gap within the object is closer to the thematic verb than
the gap within the subject, as the thematic information is coded in the participle
which occurs in sentence final position. This could potentially lead to a processing
advantage (cf. Gibson et al. 1996)
A skeptic may now interject that the differences in information structure and
verb proximity might be an alternative explanation to the subject/object asym-
metries we find. However, even though this is a logical possibility it seems highly
unlikely, in particular given a number of follow-up studies indicating that com-
plement/non-complement asymmetries persist even if the information structural po-
sition of the extraction site and the relative closeness between gap and the verb are
varied. Here is a preview of pertinent findings from our follow-up experiments. The
reader is referred to the respective sections for more detailed discussions:
? was-f?ur split out of indirect objects and internal subjects in passivized ditran-
sitive constructions again confirms a complement/noncomplement asymmetry.
While there is a Word Order main effect , i.e. the order DAT > NOM is pre-
ferred compared to NOM > DAT, there is no Word Order x Extraction
interaction effect. In other words, only the structural properties of the extrac-
tion site affects the acceptability. Information structure and verb proximity
to the gap have no effect: section 3.3
? Extraction out of subjects of unaccusatives is preferred to extraction out of
subjects of intransitive unergatives. No obvious information structural differ-
ences between these two subjects can be detected: section 3.4.2
? Melting effects could not be replicated experimentally. As discussed in section
3.1.3.2, was-f?ur split out of external subjects is sometimes claimed to improve
by scrambling the object across the subject. M?uller (2010) analyses this as
a grammatical effect. It could also be argued that scrambling the object
69
alters the information structure such that the subject becomes more ?focusy?.
However, we could not confirm the existence of melting effects experimentally:
section 3.5
? Wh-extractionout ofnon-finite sentential arguments also shows subject/object
asymmetries. Extraposing the sentential argument to a sentence-final position
has no effect on the acceptability of the extraction. More technically, there is
no interaction effect between the factors Extraposition and Extraction,
even though extraposition normally places the sentential argument in a focus
position: section 4.3
? Replacing the auxiliary in the V2 position with a main verb inverts the relative
distance between the subject andthegapandtheobject andthe gap. However,
the acceptability pattern of subextraction is not affected in any way as a recent
study suggests (data analysis in progress).
Given the vast empirical evidence, we will conclude that the island effects presented
in this section and indeed throughout this dissertation deserve a grammatical ac-
count. Let us now turn to another issue, one that is commonly brought up when
gradient data is subjected to an explanation in terms of a grammatical constraint.
3.2.5.2 Why are things as bad as they are but not worse?
We have made the case thus far that German does show CED effects. But an
elephant is in the room. The CED has always been conceived as a binary constraint,
so why is it the case that some speakers accept CED violations at all? In other words,
why are some CED violations as bad as they are but not worse?
We already mentioned in section 2.2.4 one position that often seems to be
assumed in the theoretical literature andis made explicit by Haider (1993). The logic
goes as follows: a single instance of a violation of constraint X proves that constraint
70
X does not exist. This is an assumption we have already rejected. It presupposes a
specific relation between grammaticality and acceptability, namely that violating a
grammatical constraints always and necessarily leads to strong unacceptability. This
is the necessary conclusion under the hypothesis that the grammar is strictly binary
and the assumption that all gradience we observe stems from extra-grammatical
factors.
However, where the gradience comes from is ultimately an empirical question.
There is a tradition of a binary grammar, but there are also fully worked out the-
oretical models of gradient grammars (e.g. Keller 2000, Featherston 2005). The
crucial point is that it is not a contradiction in any way that a sentence violates a
grammatical constraint but still has a status of intermediate acceptability, rather
than strong unacceptability. We simply do not know how a grammatical violation
translates into an acceptability judgment.21 Furthermore, we find that the two
grammatical constraints, the CED on the one hand and freezing on the other hand,
each individually decrease the acceptability of an example. If we put them together,
the two violations cause a stronger decrease in acceptability than either one of them
individually.
In interpreting the data we will proceed as follows: If we observe a difference
in acceptability between two conditions in a carefully controlled experiment, and if
we can beyond a reasonable doubt exclude extra-grammatical (parsing, information-
structure) explanations for this difference, we will conclude that the difference in
acceptability is caused by an underlying grammatical constraint. Of course, it is
always an option to question whether all possible extra-grammatical factors have
21It would be bizarre, of course, to postulate the violation of a grammatical constraint if we see
no decrease in acceptability whatsoever. We never find this state of affairs in our experiments.
That said, there are of course well-known cases of grammatical illusion, i.e. strings that are judged
acceptable by speakers even in controlled experiments but turn out be meaningless upon further
reflection. A famous case are comparative constructions like (i) More people have been to Russia
than I have.. We will not consider such cases in our discussion (see Phillips et al. to appear for
a general overview of ?grammatical illusions? and Wellwood et al. 2009 for a lucid discussion of
examples like (i))
71
been explored. Here explanatory parsimony comes into the picture as well. If a
grammatical constraint X is well-known to hold in a large majority of languages
and constructions, and if we observe an acceptability pattern predicted by X, it
seems reasonable that this pattern is due to X. This is the much simpler conclusion
as compared to arguing that, even though we see an acceptability pattern compatible
with X, the difference in acceptability has another source and its correlation with
X is purely coincidental. The burden of proof is on the skeptic.22
3.3 Experiment 2 - Passivized Ditransitives
3.3.1 Background
Let us turn to our next experiment, which investigates the same questions looking
at a different construction. Passivized ditransitives in German have the interesting
property of surfacing with the unmarked word order indirect object (IO) > subject.
Lenerz (1977:116) argues that the intransitive object needs to be contrastively fo-
cused in (20-b), while there is no such restriction for (20-a). Furthermore, only the
word order IO > SUB can be felicitously uttered in an out-of-the-blue context, e.g.
as an answer to the question What happened? .
(20) a. Ich
I
wei?,
know
dass
that
einem
a.DAT
Professor
professor
ein
a.NOM
Student
student
vorgestellt
introduced
wurde.
was
?I know that a student was introduced to a professor.?
b. Ich
I
wei?,
know
dass
that
ein
a.NOM
Student
student
einem
a.DAT
Professor
professor
vorgestellt
introduced
wurde.
was
?I know that a student was introduced to a PROFESSOR (not to the
dean).?
Sabel (1999: 7ff.) convincingly argues that the nominative subject stays inside the
22This is essentially, albeit phrased in slightly different terms, the same conclusion reached by
Featherston (2005a,b) for the that-trace effect and Superiority in German.
72
VP. He proposes the following structure:
(21) VP
einem.dat Professor VP
ein.nom Student V
Sabel argues based control and binding data that the nominative marked NP can
stay in its in-situ position in German, i.e. as the complement of V. Recall from
our discussion in section 3.1.2.3 that, unlike English, German does not obligatorily
require internal arguments to raise to SpecTP to receive nominative case. Case
assignment can take place in-situ (see Wurmbrand 2006 for a proposal of the mech-
anism of case assignment in German). Sabel assumes the IO to be adjoined to VP
but see the discussion section below for alternative proposals.
We will compare extraction out of in-situ subjects and indirect objects. Fur-
thermore, we will manipulate the word order to see whether the acceptability of
subextraction out of an argument is affected by the position of this constituent
relative to the other argument. In particular, this allows us to test a potential al-
ternative analysis of the findings in 3.2, in terms of information structural role or
verb proximity. Let us first take a look at the stimuli before considering the various
predictions
3.3.2 Methodology and Design
23 native speakers of German participated in this online study. The methodological
protocol laid out in section 3.2 was followed. The study test was-f?ur split , manip-
ulating the factors Word Order and Sub/IO. Two baseline conditions without
extraction were added. This yields the following set of stimuli.23
23The experiment in fact also included two Across-the-Board was-f?ur split conditions, i.e. 8
conditions total. We will only present the first 6 conditions here and discuss ATB-was-f?ur split in
73
(22) a. Gestern
yesterday
wurde
was
einem
a.DAT
Professor
professor
ein
a.NOM
Student
student
vorgestellt.
introduced
?Yesterday a student was introduced to a professor.? [O>S, -ext]
b. Gestern
yesterday
wurde
was
ein
a.NOM
Student
student
einem
a.DAT
Professor
professor
vorgestellt.
introduced
?Yesterday it was a professor a student was introduced to.? [S>O, -ext]
c. Was
what
wurde
was
denn
PRT
einem
a.DAT
Professor
professor
f?ur
for
ein
a.NOM
Student
student
vorgestellt?
introduced
?What kind of student was a professor introduced to?? [O>S[t]]
d. Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
Professor
professor
ein
a.NOM
Student
student
vorgestellt?
introduced
?What kind of professor was a student introduced to?? [O[t]>S]
e. Was
what
wurde
was
denn
PRT
f?ur
for
ein
a.NOM
Student
student
einem
a.DAT
Professor
professor
vorgestellt?
introduced
?What kind of student was a professor introduced to?? [S[t]>O]
f. Was
what
wurde
was
denn
PRT
ein
a.NOM
Student
student
f?ur
for
einem
a.DAT
Professor
professor
vorgestellt?
introduced
?What kind of professor was a student introduced to?? [S>O[t]]
3 lexicalizations for each condition were created, yielding a total of 108 items, which
were grouped in six Latin-squared lists. The resulting 18 stimuli together with 43
items from different experiments and 23 fillers were presented in pseudo-randomized
order.
Only indefinite animate NPs were used for both subjects and indirect objects
in all conditions. Again the modal particle ?denn? was used to demarcate the vP
-boundary. (22-a) and (22-b) serve as base line conditions to gauge how much the
marked word order influences the acceptability. In (22-c)was-f?ur split takes places
from the internal subject position, and in (22-d) the split originates from the IO
position, in both cases with the unmarked word order IO > SUB. In (22-e) and
(22-f) we again extract out of the internal subject and the IO, this time using the
section 3.7. See the appendix for the full list of items.
74
marked word order SUB > IO.
3.3.3 Results
We found that there is a marginally significant main effect for the factor Word
Order, i.e. speakers slightly prefer the unmarked word order IO > SUB over the
marked word order SUB > IO. This does not considerably affect the acceptability of
the no-extraction condition, (22-b), which is still given a very high rating (5.68). A
pairwise comparision shows that the difference between the baseline conditions is not
significant (t(1,68)=.27, p=.13). We see Extraction main effects for both was-f?ur
split out of subjects and out of IOs, but there is a significant asymmetry between the
two. While was-f?ur split out of the subject, (22-c), still yields intermediate ratings
(3.57), extraction out of IOs, (22-d), receives a much lower rating (2.01) (pairwise
comparison: t(1,68)=5.33, p<.001). Combing extraction and marked word order
further decreases the acceptability (3.06 vs. 1.72, (t(1,68)=5.12, p<.001). It is
interesting to note, however, that the effects do not seem to be cumulative in this
case in that there is no interaction effect for Sub*Marked or IO*Marked. In
other words, there is a difference in absolute numbers between extraction out of
subjects and IOs with SUB > IO word order compared to IO > SUB order, but
this difference stems solely from the marked word order main effect. The results are
summarized in Figure 3.4, the descriptive statistics in Table 3.4 and the ANOVA
results in Table 3.5.
3.3.4 Discussion
3.3.4.1 How to interpret Likert-scale results
Before we delve into the details of the analysis and the theoretical implications of
these results one methodological point needs to be clarified. It is immediately notice-
75
gid13gid1
gid14gid1
gid15gid1
gid16gid1
gid17gid1
gid18gid1
gid19gid1
gid3gid2gid7gid1gid9gid1gid5gid6gid4gid1 gid5gid6gid4gid1gid9gid1gid3gid2gid7gid1 gid3gid2gid7gid1gid9gid1gid10gid8gid1gid5gid6gid4gid11gid1 gid10gid8gid1gid3gid2gid7gid11gid1gid9gid1gid5gid6gid4gid1 gid10gid8gid1gid5gid6gid4gid11gid1gid9gid1gid3gid2gid7gid1gid1 gid5gid6gid4gid1gid9gid1gid10gid8gid1gid3gid2gid7gid11gid1gid1
Figure 3.4: Experiment 2: Means
Table 3.4: Descriptive Statistics
Mean SD
Dat > Nom 6.00 1.73
Nom > Dat 5.68 1.66
Dat > [t Nom] 3.57 2.03
[t Dat] > Nom 2.01 1.26
[t Nom] > Dat 3.06 1.85
Nom > [t Dat] 1.72 1.11
Table 3.5: ANOVA
F p
Sub 76.1 ***
IO 330.1 ***
Sub*IO 166.1 ***
Scr 8.9 **
Sub*Scr .1 .912
IO*Scr 1.1 .317
able that the means of the was-f?ur split condition are lower compared to Experiment
1 across the board. This is a good spot to remind ourselves of a very important fact
about acceptability judgment data. We should never draw any significant conclu-
sions from the absolute rating we get for a condition. This might sound odd since
the scale is anchored, that is the various points translate to a statement about a
speaker?s intuition, e.g. rate a sentence with 7 if you find it fully acceptable. This is
certainly true, and there is a sense in which there is some meaning to the absolute
values. If, for instance, a sentence like What did you see? receives a rating around
2, we know that there is something fundamentally wrong with our experiment. At
76
the same time, however, we should not be too worried if it is judged around 5 on a
7-point scale. Many factors influence the absolute ratings.
One major factor are the other items that appear in the questionnaire (fillers
and items from other experiments), which strongly bias the ratings. People are
affected by the context in which they are asked to give a judgment. A sentence
like What did you see? is likely to get a rating higher than 6 if the majority of the
items in the same questionnaire include strong violations (strong islands, agreement
mismatches, etc.). The same sentence might be rated only around 5 if the majority
of fillers are simple monoclausal declaratives that do not contain any violations.
It also seems to be a fact about Likert scales in general and not particular
to acceptability judgments that many participants have a tendency to not use the
extremes of the scale (i.e. giving 1s or 7s). This might be related to the eagerness of
some participants to leave themselves the option of giving even stronger ratings in
case of even more extreme items. These points make it methodologically question-
able to compare ratings across various experiments. Too many factors could differ,
above all the fillers and the participants.
The bottom line is that substantial claims can only be made about the rel-
ative difference between two conditions that differ minimally in the effect under
investigation and appear in the same questionnaire. We are concerned with relative
judgments exclusively, not absolute judgments. It does not make sense to say: a
sentence is ungrammatical below a value X , above X it is grammatical.
3.3.4.2 Analysis of the effects
Let us return to our experiment. We saw above that all NPs in the stimuli were
controlled for animacy and definiteness. This is crucial to preclude word order biases
that are well known to be triggered by these factors (see Lenerz 1977, M?uller 1999
among others). However, definiteness especially can play a key role in the overall
77
decreased acceptability of the was-f?ur split conditions. It seems that, for reasons
that are not totally lucid, was-f?ur split is preferred when the other NP in the clause
is definite. This, however, affects all subextraction conditions across the board and
does not bias one condition disproportionally.
With all of these caveats in mind, let us look at our findings. There is a
clear asymmetry in acceptability between was-f?ur split out of subjects and indirect
objects. This is not a surprise if Sabel?s (1999) structure, repeated here, is correct:
(23) VP
IO VP
sub V
The nominative marked NP, the internal subject, is in the complement position of
V and as such expected to allow extraction. Our findings also confirm Den Besten?s
(1985) original insight that IOs do not allow subextraction. The isolated examples
in Bayer et al. (2001) and Lutz (2001) (cf. the discussion section 3.1.3) strongly
diverge from the means and are clearly a minority view. It is very unlikely that
the mean is indicative of a cleft between the speakers, with some accepting the
extraction while others do not.24 This is illustrated by the histograms in Figure 3.5.
The histogram on the left represents all ratings for subextraction out of IO with the
IO > SUB order, and the chart on the right with the SUB > IO. Bayer et al.?s and
Lutz?s judgments are certainly outliers.
24Note that we did not test the exact examples Bayer et al.?s and Lutz?s used. While their
examples were in active voice, we used passives. Melchiors (2007) looked at active sentences and
found a similar picture: extraction out of indirect objects is strongly degraded, in contrast to
Bayer et al.?s and Lutz?s judgments. This makes it unlikely that passive voice and the availability
of extraction out of indirect objects interacts in some way. What is more plausible is that passive
voice has a certain effect on the judgments across the board. as we will see in Experiment 3 below.
We do not expect any more or less severed effect on indirect objects.
78
Figure 3.5: Subextraction out of IOs - individual distribution
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3
gid10gid20gid1gid3gid2gid9gid11gid1gid7gid8gid6
gid3
gid8
gid4gid3
gid4gid8
gid5gid3
gid5gid8
gid6gid3
gid4gid19
gid14gid18
gid21gid14
gid16gid13
gid22
gid16gid22gid18gid2gid1gid12gid19gid23gid2gid1gid11gid1gid4gid2gid5gid8gid10gid10gid4
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3
gid7gid8gid6gid1gid10gid20gid1gid3gid2gid9gid11
gid3
gid4gid3
gid5gid3
gid6gid3
gid7gid3
gid4gid19
gid14gid18
gid21gid14
gid16gid13
gid22
gid17gid23gid19gid2gid1gid13gid20gid24gid2gid1gid12gid1gid4gid2gid4gid3gid11gid11
What does the CED have to say about the islandhood of IOs? If we assume
that DOs are merged in a complement position, no matter which other position IOs
are merged in they would always constitute opaque domains, irrespective whether
they are analysed as adjuncts to VP or specifiers of some sort. Our results are
compatible with a structure such as the one given initially by M?uller (1995: 186),
assuming as we have been so far that nominative marked themes in passives occupy
the same position as DOs:
(24) VP
IO ?V
DO V
Grewendorf (1988), however, argues based on the following binding data that the
DO asymmetrically c-commands the IO:
79
(25) a. Der
the
Arzt
doctor
zeigte
showed
den
the
Patientenj
patient.acc
sichj/*ihnj
himself/him
im
in.the
Spiegel
mirror
b. Der
the
Arzt
doctor
zeigte
showed
dem
the
Patientenj
patient.DAT
ihnj/*sichj
him/himself
im
in.the
Spiegel
mirror
The anaphor in the IO position can be bound by the DO but not vice-versa. This
data seems to suggest a structure like (26):
(26) VP
DO ?V
IO V
M?uller acknowledges Grewendorf?s data and tries to reconcile (24) and (26) by
assuming that the IO is asymmetrically c-command in its based position by the DO
but the IO moves across to DO to a higher position to get case, as illustrated in
(27):
(27) VP
IO+case VP
DO V ?
<IO?case> V?
PP-arguments V
This structure is strikingly similar to Larson?s (1988) analysis of double object con-
struction in English and Baker?s (1988) UTAH. If that is the correct structure the
CED would predict both IOs and DOs to be islands. Perhaps this structure is
compatible with our results, as both extraction out of IOs and internal arguments
80
are degraded to a certain extent. The asymmetry between the two could then be
construed as a freezing effect, as the IO raises overtly for case reasons.
However, I do not think that this is the right conclusion to draw. First,
Grewendorf?s data that prompted a structure such as (27) does not seem to hold
up. Featherston (2002) runs a detailled acceptability judgment study indicating that
Grewendorf?s binding data does not hold up. Speakers judge reflexives more accept-
able than pronouns across the board with both dative and accusative antecedents. I
personally do not share Grewendorf?s judgments in (25) either. In addition, Feath-
erston reports the same effect in word order preference IO > DO vs. DO > IO
we also found. We will not try to give a full-fledged account of German binding
here. What is crucial is that the key data in favor of an hierarchy where the DO
asymmetrically c-commands the IO does not seem to hold empirically.25
Our results might be be interpreted as being reflective of a structure such as
(27). In that case, extraction out of the subject would be a mere CED violation
while extracion out of the IO would be a CED and a freezing violation. However,
we saw above that absolute ratings should generally not be used to draw conclusions
of any kind. Only relative judgments can be interpreted in a meaningful way. As
such, the fact that extraction out of internal arguments is given a score of 3.57 does
not imply that any grammatical principle is violated. The general cost incured by
was-f?ur split combined with the cost of passivisation is more likely to be the culprit
for the decreased score. As such, our results are fully compatible with a simpler
structure such as (24).
Ultimately, these cases raisequestions aboutthe thematichierarchy, i.e. whether
DOs or IOs are closer to the verb. Our results point to the conclusion that DOs are
complements of the verb, whereas IO are in some higher non-complement position.
25The relevance of the binding data is theory dependent, to begin with, as Norbert Hornstein
(p.c.) points out. In a theory such as Reinhardt & Reuland (1993), it is not clear that the
c-command relations play any role for establishing licit binding relations.
81
Another interesting follow-up question arises for ?persuade? type constructions which
both take a DO and a CP-complement. Does the DO allow subextraction even when
a CP-complement is present or does the CP occupy the complement position of V,
which results in the DO showing CED effects? The relevant comparision would be
the following:26
(28) a. ?Was
what
hat
has
der
the
Peter
Peter
f?ur
for
einen
a.ACC
Mann
man
?uberredet
persuaded
einen
a
Marathon
marathon
zu
to
laufen?
run
?What kind of man did Peter persuade to run a marathon??
b. Was
what
hat
has
der
the
Peter
Peter
f?ur
for
einen
a.ACC
Mann
man
gesehen?
seen
?What kind of man did Peter see?
To my ear, (28-a) sounds a little degraded compared to (28-b). This would suggest
that the CP in fact occupies the complement position of V and the DO is located in
some other non-complement position. The contrast in acceptability, however, is not
robust enough to jump to any conclusions based on it. A controlled acceptability
judgment study would be needed, a task left for future research.
Let us now have a closer look at extraction out of internal subjects. While it
is significantly more acceptable than extraction out of indirect objects, speakers do
not seem to behave uniformly. This is illustrated in Figure 3.6.
The chart on the left indicates the distribution of subextraction out of internal
subjects with a IO > SUB word order and the chart on the right with a SUB > IO
word order. Let us first focus on the former. The distribution approximates bimodal
26The same kind of comparison can be constructed for English too: (i) ?Who did you persuade
a friend of to run a marathon? vs. (ii) ?Who did you meet a friend of?. The judgments are not
crystal-clear. Extraction out of animate DPs is already degraded in English for many speakers,
but it is conceivable that we would still find a further contrast between (i) and (ii) on top of
that. A relatively straightforward experiment could give an answer. Since there are just too many
relatively straightforward (and less straightforward) experiments to run, I will have to leave this
open for future research.
82
Figure 3.6: Individual distribution
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid3gid2gid9gid1gid10gid20gid1gid7gid8gid6gid11
gid3
gid6
gid9
gid11
gid4gid5
gid4gid8
gid4gid19
gid14gid18
gid21gid14
gid16gid13
gid22
gid14gid20gid18gid22gid1gid12gid1gid6gid2gid8gid9gid8gid5
gid17gid23gid19gid2gid1gid13gid20gid24gid2gid1gid12gid1gid5gid2gid3gid6gid5gid9gid9
gid15gid1gid12gid1gid9gid11
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid10gid20gid1gid7gid8gid6gid11gid1gid3gid2gid9
gid3
gid8
gid4gid3
gid4gid8
gid5gid3
gid4gid19
gid14gid18
gid21gid14
gid16gid13
gid22
gid18gid24gid20gid2gid1gid14gid21gid25gid2gid1gid13gid1gid4gid2gid11gid8gid7gid4gid5
behavior. Speakers both reject and accept the extraction a good percentage of the
time. There could be a number of reasons for this trend. First, as mentioned above,
speakers tend to disprefer was-f?ur split in contexts where all NPs in the clause
are indefinites. This is very unlikely to stem from a grammatical constraint but
seems more likely to have a motivation rooted in information structure. Informally
speaking, it might be somewhat odd to inquire about the kind of some NP while not
singling out another participant of the event. You can mimic this effect in English:
(29) What kind of man did a woman meet?
While (29) is certainly not ungrammatical, there is some degree of oddity involved.
Different speakers are likely to be more or less skillful in creating a plausible prag-
matic situation for such an utterance, and hence some might reject it for that reason.
Second, there are two potential structures that could be assigned to (22-c).
83
(30) a. dass [VP einem Professor [V ? ein Student vorgestellt]] wurde
b. dass [TP einem Professori [TP ein Studentj] [V P ti tj vorgestellt]] wurde
In (30-a) both the internal subject and the indirect object are in-situ. In (30-b) the
subject is moved to the derived subject position in SpecTP and the indirect object
is scrambled across it, adjoining to TP. Sabel (1999), discussing this construction
in detail, proposes the structure in (30-a). He concedes, however, that there is no
way to test empirically whether the IO vacuously scrambled higher to adjoin to TP.
He discards this possibility on theoretical grounds, making the plausible minimalist
assumption that vacuous movement (i.e. movement with no effect at PF or LF) is
disallowed. It is still conceivable that speakers assigned different structures to the
same string. This would explain the bimodal distribution. For the former group,
the internal subject would remain in its in-situ position as the complement of V and
as such be transparent for extraction. For the later group, the degradation of the
example would be a result of freezing, similar to what we found in Experiment 1.
Note, however, that I am not aware of any theory that proposes this kind of vacuous
movement. It is only a theoretical possibility, albeit not a particularly plausible one.
Alternatively, it could turn out that some speakers assign these cases a rep-
resentation such as (24) and others the one in (27). Ultimately, this particular
experiment cannot give us a sufficiently satisfying answer to this question. We will
investigate extraction out of internal and external arguments in detail in Experiment
3 and 4, in sections 3.4 and 3.5, respectively.
In addition, we confirm the intuition that IO > SUB is the preferred word order
for passivized ditransitives. There is a significant Word Order main effect. This
trend also shows up in the non-extraction conditions but fails to reach significance in
the pairwise comparision. Interestingly, there is no Word Order x Extraction
interaction effect, neither for subjects nor for objects. Importantly, we find that
84
the acceptability of the extraction solely depends on the syntax of the extraction
site and cannot be explained by an account in terms of processing or information
structure. If we control for the word order main effect, the position of the gap in
the string does not affect the acceptability. Subjects and indirect objects show the
same asymmetry no matter in which linear order they surface, as indicated by the
lack of interaction effects for Sub*Scr and IO*Scr in table 3.5. If verb proximity
or topichood were the only factors that govern extractability, we would expect them
to show up in our data.27
3.3.5 Conclusion
Summarizing the findings of this chapter, we saw that extraction out of indirect
objects is strongly degraded for virtually all speakers, contrary to some claims in
the theoretical literature. This is expected from the CED if IOs are not complements
of V. Additionally, we got mixed results for extraction out of internal subjects, a
topic to which we will return promptly and in much more detail in Experiments
3 and 4. Finally, our data gives us good reasons to believe that the extraction
asymmetries we find cannot be reduced to extra-grammatical factors.
27Den Besten?s (1985) original insight that IOs do not allow was-f?ur split is confirmed. However,
in his assessment that there is an asymmetry between extracting out of in-situ vs. moved subjects,
he fails to control for the marked word order main effect . He judges examples like (22-c) as fully
acceptable while ruling out examples like (22-e). This seems like a mischaracterization of the facts,
stemming from the bias of binary categoricity and the lack of more rigid data collection.
85
3.4 Experiment 3 - Internal and external arguments
3.4.1 Experiment 3A
3.4.1.1 Introduction
This study follows up on the findings in Experiment 1 and 2 by comparing was-f?ur
split out of internal and external arguments. Concretely, we will contrast subextrac-
tion out of in-situ subjects of unergatives, unaccusatives and passives as well as out
of objects. At this point, the predictions should be clear. The CED predicts that
internal arguments, i.e. unaccusative and passive subjects and objects, should show
uniform behavior while extraction out of unergative subjects should be degraded.
This is roughly the picture that emerges.
3.4.1.2 Methodology and Design
37 native speakers of German (by self-assessment) with no prior training in linguis-
tics, predominately speakers of Austrian German, were asked to rate sentences on a
7-point Likert scale presented to them in an online questionnaire. The data of all 37
participants was used for data analysis. The same methodological protocol outlined
in 2.2.2 and in the previous experiments was followed.
The experiment had a 5x2 structure with the independent variables Argu-
ment Type and Extraction. We constructed examples of was-f?ur split out of
subjects of unaccusatives, passives, and objects as well as out of the intransitive and
transitive unergative subjects. For each predicate type we added a control condition
where no split takes places. Again, an adverbial marker was used to demarcate the
vP boundary. This results in 10 conditions. The following shows the 5 +extraction
conditions, the control conditions are parallel with the entire was-f?ur phrase fronted.
86
(31) a. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
ein
a
Mann
man
gestern
yesterday
am
on.the
Nachmittag
afternoon
erschienen
appeared
ist.
is.
?I wonder what kind of man appeared yesterday afternoon.? [unacc]
b. Ich
I
frage
ask
mich,
myself
was
what
der
the
Mann
man
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
einen
a
Piloten
pilot
gesehen
seen
hat.
has
?I wonder what kind of pilot the man saw yesterday afternoon.? [obj]
c. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
Mann
man
gesehen
seen
wurde.
was
?I wonder what kind of man was seen yesterday afternoon.? [pass]
d. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
Mann
man
den
the
Piloten
pilot
gesehen
seen
hat
has
?I wonder whatkind of man saw the pilot yesterday afternoon.? [trans]
e. Ich
I
frage
ask
mich,
myself
was
what
f?ur
yesterday
ein
on.the
Mann
afternoon
gestern
for
am
a
Nachmittag
man
geschlafen
slept
hat.
has
?I wonder what kind of man slept yesterday afternoon.? [unerg]
3 lexicalizations of every condition were created. The same verbs were used for
the transitive, object and passive conditions. Different verbs had to be used in the
unergative and unaccusative conditions. Note that I selected the unergative and
unaccusative predicates solely based on which auxiliary they selected. This might
have been a too simplistic metric, even more so, in light of Sorace?s (2000) discussion
of the unaccusative-unergative continuum. The distinction between the two types
does not seem fully binary, but there is a sense in which one verb can be more
unaccusative than another one but still less unaccusative than a third one. We will
return to this issue in the discussion section below, and we will conduct a follow-up
study using only extreme cases on the scale in Experiment 3B below.
87
3.4.1.3 Results
Our results indicate that two things matter: whether a predicate has one or two
arguments and whether the extraction originates from an internal or external ar-
gument position. Extraction out of objects, (31-b), is preferred to extraction out
of transitive subjects, (31-d) (4.36 vs. 3.51, t(1,110)=3.49, p<.001). Objects and
transitives have the lowest values for the -extraction control condition. This seems
to be a direct reflection of the fact that these are the conditions with two overt
arguments. The fact that an additional ?-role needs to be assigned is burdensome
to the parser and causes a decrease in acceptability. Unaccusatives, passives and
unergatives show virtually identical values for the control condition.
What about the effect of was-f?ur split ? The best cases of extraction are
out of unaccusatives and passives. In both cases the extraction originates from
an internal argument position in a configuration where this is the only argument
present. For unergatives, which also only have one argument, extraction is slightly
degraded. This seems to be a reflection of the fact that the extraction originates
from an external argument position. Turning to the two conditions with two overt
arguments, we observe a contrast between objects and transitives. This seems to
be a reflection of the fact that in the object case we are extracting from an internal
argument position while we are extraction from an external argument position in
the transitive condition. See also the descriptive results in Table 3.6.
Table 3.6: Experiment 3: Descriptive Statistics
no Split Split
Unaccusative 5.79 4.65
Object 5.39 4.36
Passive 5.84 4.74
Transitive 5.52 3.51
Unergative 5.85 4.32
88
Let us now turn to the chart in Figure 3.7. For reasons of better legibility, we
zoomed in on the area between 3 and 6 on the scale. The two top grey lines for
unaccusatives and passives are almost identical. The grey dotted object line shows
an overall lower acceptability, an Argument Type main effect, but is still parallel
to the unaccusative and the passive lines, i.e. there is no interaction effect. The two
black lines represent those conditions where the extraction originated in an external
position. The solid black line, the transitive condition in (31-d) , shows the steepest
slant, which points to an interaction effect between the factors Argument Type
x Extraction. The dashed black line, the unergative condition in (31-e), is also
non-parallel to the unaccusative, passive and object lines, but its slant is slightly
less steep.
gid21gid17gid20gid20gid1
gid21gid17gid23gid20gid1
gid22gid17gid20gid20gid1
gid22gid17gid23gid20gid1
gid23gid17gid20gid20gid1
gid23gid17gid23gid20gid1
gid24gid17gid20gid20gid1
gid18gid9gid16gid15gid1 gid25gid9gid16gid15gid1
gid5gid12gid6gid8gid8gid1
gid5gid12gid9gid13gid10gid1
gid4gid13gid6gid12gid14gid1
gid3gid6gid14gid14gid1
gid2gid7gid11gid1
Figure 3.7: Experiment 3A: Results
So far we mostly considered the descriptive statistics, which gave us a first
picture of the results. Let us turn to the statistical tests now. We conducted 2x2
ANOVA subanalyses of the Argument Type x Extraction interaction effects
89
The results are summarized in Table 3.7.
Table 3.7: ANOVA 2x2 subanalyses ArgType x Ext interaction effects (p-values)
Unaccusative Unergative Transitive Passive Object
Unaccusative N/A .26 ?? 1 .68
Unergative .26 N/A ? .18 ?
Transitive ?? ? N/A ??? ???
Passive 1 .18 ??? N/A .63
Object .68 ? ??? .63 N/A
We observe that the transitive condition shows interaction effects with all
other conditions. This means that subextraction out of transitive subjects incurred
a disproportionally higher cost than any other extraction. Recall that the transitive
condition is the only condition with both a more complicated argument structure
(2 arguments instead of 1) and where the extraction originated from an external
argument position.
The unergative condition only shows marginal interaction effects with transi-
tives and objects but not with unaccusatives and passives. While we saw a slight
trend in the descriptive statistics above that unaccusative and passive subjects, i.e.
internal arguments, tolerate extraction better than unergatives, i.e. external argu-
ments, this trend fails to reach significance. Since we do see a pattern in the direction
we expect we will follow-up on the contrast between unergatives and unaccusatives
in Experiment 3B below.
Passives, unaccusatives and unergatives do not show any Argument Type
x Extraction interaction effects with respect to each other, i.e. extraction causes
the same relative degradation in all three cases. The last interesting aspect of the
results we would like to stress is that unergatives and transitives show an effect. This
suggests that even though we are extracting out of external argument positions in
both cases, the more complicated argument structure in the transitive case has a
disproportionally stronger effect on the extraction. This is a similar effect to what
90
Kravtchenko et. al. (2009) found for extraction out of Russian subjects.
3.4.1.4 Discussion
The results of thisexperiment continue the trendwe saw inthe previous experiments.
We see evidence pointing to the conclusion that the CED holds in German: subjects
merged as specifiers are degraded as compared to subjects merged as complements
and objects. In other words, the asymmetry we saw in unergative subjects vs.
objects seems to extend to internal arguments in general. This does not come as
a surprise. Recall from the discussion in section 3.1.1 that the term ?subject? only
plays a descriptive role but no constraints are defined by it. The only notion that
matters is being in a sister relationship with V. Subjects of unaccusatives, passivized
subjects and objects all occupy this structural position and thus allow extraction.
Subject of unergatives are merged as specifiers, and subextraction out of them is
degraded.
Recall, however, thatintheone-argument conditions(unergatives, unaccusatives
and passives), we saw a slight trend disfavoring unergatives, but this trend did not
reach significance in the statistical test. One potential source of this lack of an effect
could be the choice of unergative and unaccusative verbs. I based my choice solely
on the auxiliary selection, which seems to have been a too simplistic metric (see
Appendix B for a full list of verbs used). Sorace (2000) provides evidence for the ex-
istence of an unaccusative-unergative continuum, rather than a binary distinction.
In Experiment 3B below we will compare the two types, choosing only the most
extreme cases on the scale.
Furthermore, note that this is evidence that internal arguments can stay in-
situ, no matter if they are assigned nominative or accusative case. In this regard
German contrasts with English, which forces the movement to SpecTP of subjects
of unergatives, unaccusatives and passives alike. Our results lead us to take sides
91
with claims from the theoretical literature, according to which the EPP does not
hold in German (Haider 1993, Rosengren 2002). Nominative case assignment can
take place in-situ in German (Wurmbrand 2006).28
These results also shed more light on some of the loose ends of the result of
the interpretations in Experiment 2. Remember that extraction out of passivized
subjects patterns in a roughly bimodal way. One potential explanation we consid-
ered was the fact that our example allowed two structural analyses, one in which the
internal subject stays in-situ and one in which it moves to SpecTP. In the latter case
degradation would be expected as a result of freezing. In this experiment, however,
an example like (31-c) forces a construal of the subject in its in-situ position. Hence
no freezing effects occur.
In addition to the interaction effect we found for unergatives as compared to
all other conditions, we also find an Argument Type main effect for unergatives
and objects. The effect is most plausibly due to the extra argument in those con-
ditions. It is well known that additional referents imply higher parsing cost, which
in turn results in a lower acceptability rating. This is an effect fully independent of
extraction.
We conclude that this experiment fits in with the trends we saw in our previous
studies. There is, however, again the concern that extra-grammatical factors are
(partially) responsible for the pattern we see. After all, we saw that the number of
arguments has an effect on the acceptability of the extraction. In order to exclude
the possibility that effects we saw can be reduced to this difference, we conducted a
follow-up experiment where we contrasted extraction out of subjects of intransitive
unergatives vs. unaccusatives.
28We are agnostic as to whether this is instantiated as a long distance Agree operation or as
covert raising.
92
3.4.2 Experiment 3B
3.4.2.1 Methodology and Design
37 native speakers of German (by self-assessment) with no prior training in linguis-
tics (a different set of speakers from Experiment 3A), were asked to rate sentences
on a 7-point Likert scale presented to them in an online questionnaire. The data of
all 37 participants was used for data analysis. The same methodological protocol
outlined in 2.2.2 and in the previous experiments was followed.
The experiment had a 2x2 structure with the independent variables Argu-
ment Type and Extraction. We constructed examples of was-f?ur split out of
subjects of unaccusatives and intransitive unergatives. This design minimizes the
extra-grammatical factors discussed above. Again, an adverbial marker was used to
demarcate the vP boundary. This gives us the following 4 conditions:
(32) a. Was
what
f?ur
for
M?anner
men
sind
are
denn
PRT
am
on.the
Nachmittag
afternoon
angekommen?
arrived
?What kind of men arrived yesterday afternoon?? [unacc, -ext]
b. Was
what
sind
are
denn
PRT
f?ur
for
M?anner
men
am
on.the
Nachmittag
afternoon
angekommen?
arrived
?What kind of men arrived yesterday afternoon?? [unacc, +ext]
c. Was
what
f?ur
for
M?anner
men
haben
have
denn
PRT
am
on.the
Nachmittag
afternoon
gearbeitet?
worked
?What kind of men worked yesterday afternoon?? [unerg, -ext]
d. Was
what
haben
have
denn
PRT
f?ur
for
M?anner
men
am
on.the
Nachmittag
afternoon
gearbeitet?
worked
?What kind of men worked yesterday afternoon?? [unerg, +ext]
The verbs used were based on the unaccusativity hierarchy proposed by Sorace
(2000). To bring out the unergative-unaccusative contrast the strongest, only words
from either end of the scale were chosen (change of states verbs on the unaccusative
end, and controlled process (non motional) verbs on the unergative side). See the
appendix for the full set of items.
93
gid21gid17gid20gid20gid1
gid22gid17gid20gid20gid1
gid23gid17gid20gid20gid1
gid24gid17gid20gid20gid1
gid25gid17gid20gid20gid1
gid26gid17gid20gid20gid1
gid27gid17gid20gid20gid1
gid18gid5gid15gid10gid1 gid28gid5gid15gid10gid1
gid2gid7gid3gid4gid4gid13gid9gid3gid11gid14gid5gid1
gid2gid7gid5gid8gid6gid3gid11gid14gid5gid1
Figure 3.8: Experiment 3B: Results
3.4.2.2 Results
We see the familiar contrast between unergatives and unaccusatives, as indicated
by the non-parallel lines in Figure 3.8. Again extraction out of unergative subjects
(4.67) is degraded compared to extraction out of unaccusatives (5.59) (t(1,81)=3.05,
p=.0013). The base line conditions without extraction are almost identical (5.85.
for unergatives vs. 5.95 for unaccusatives, t=.54)). An ANOVA confirms that there
is a Predicate Type x Extraction interaction effect: F(1,80)=74.714, p<.001).
3.4.2.3 Discussion
This experiment confirms the results found in the previous study, while excluding
the potential extra-grammatical factors. We have two virtually identical sets of
strings that should show no differences in parsability and information structure.
A. Goldberg (p.c.) argues that there might be a difference between unaccusatives
and unergatives in how easily the subject can be focused. She points out that the
94
subject can be stressed contrastively more easily in (33-a) compared to (33-b). This
would account for the different extraction patterns, she argues, as focus domains
allow extraction more easily (capitals indicate prosodic stress).
(33) a. The MAN arrived.
b. ??The MAN worked.
Goldberg?s observation for (33) seems empirically correct and can be replicated for
German. However, there are two issues with her objection. On the theoretical side,
it is not clear at all why these information structural facts should affect syntactic
extraction. For a number of counterarguments against her proposal see section
3.2.5.1 and Lidz & Williams (2009).
On the empirical side, focusability of the subject is not coextensive with un-
accusativity. There are many unaccusative verbs that behave just like unergatives
in this respect.29
(34) ??The MAN died.
It seems that the extraction patterns for these verbs are the same as the one found
in our experiment:
(35) a. Was
what
sind
are
denn
PRT
f?ur
for
M?adchen
girls
rot
red
geworden?
turned
?What kind of girls blushed??
b. ?Was
what
haben
have
denn
PRT
f?ur
for
M?adchen
girls
gearbeitet?
worked
?What kind of girls worked??
A further study would have to be conducted to confirm this intuition. As those
types of preciates that facilitate the focusability of the subject cross-cut the unerga-
29Thanks to Alexander Williams for bringing my attention to this point.
95
tive/unaccusative divide, it would be of particular interest to test whether they are
two separate factors that affect the acceptability of the extraction. First, we would
need to establish in an independent study which verbs most easily allow focus inter-
pretations of the subject. This is likely to result in a continuum of verbs. For now,
let us call those verbs that most easily allow focused subjects, +focus and those
that do not -focus. A further study could then combine these factors and look at
the acceptability of extraction out of +focus, unaccusatives, -focus, unaccusatives,
+focus, unergatives and -focus, unergatives. No matter what the outcome of such
a study would be, it would have very interesting consequences for our theories of
extraction. If there is an unaccusative/unergative divide on top of the focus effect
- which I take to be the most likely result - it seems that a grammatical constraint
and an information structural constraint are needed to explain the data. If we do
not see any unaccusative/unergative difference but only a focus effect, a purely in-
formation structural account might be feasible. Then the question arrises what the
details of this account would be and how it would handle the problems raised by
Lidz & Williams (2009) and in section 3.2.5.1. We leave these questions open for
future research.
We conclude this section with another small remark on the role of unaccusativ-
ity with respect to subextraction. It is well known that not all unaccusatives are the
same. As mentioned above, Sorace (2000) and Keller (2000) report cross-linguistic
evidence drawn from auxiliary selection for an unergative-unaccusative continuum.
It would be interesting to test experimentally whether auxiliary selection as an indi-
cator of where a specific verb is situated on the unergative-unaccusative continuum
correlates with the permeability of the subject. Here French presents an interesting
case. Certain verbs in French optionally take either a ?have? or ?be?. The combien de
split-construction seems to be sensitive to this contrast (M. Gagnon?s judgment):
96
(36) a. Combieni
how-many
sont
are
apparus
appeared
[ti des
of.the
etudiants]?
students
b. ?Combieni
how-many
ont
have
apparus
appeared
[ti des
of.the
etudiants]?
students
?How many students have appeared??
In both cases combien is subextracted from a post-verbal position. The choice of
the auxiliary, however, causes an asymmetry between the two examples. Native
speakers have the intuition that the subject in (36-b) is somewhat more agentive
than in (36-a). (36-a) is in a certain sense more unaccusative than (36-b) and
allows extraction more easily. We leave it to future work to further investigate
minimal pairs of that form cross-linguistically to see how notions like unaccusativity,
agentivity, auxiliary selection and permeability of the subject are interrelated.
3.4.3 Conclusion
Wrapping up this section, we can conclude that the experiments in this section
provide further evidence for the existence of CED effects in German. We saw that
controlled experimental elicitation of the data can bring out these effects, which were
only partially noticed in the theoretical literature. We tested complement/non-
complement asymmetries in a number of different contexts and have made every
effort to exclude the possibility of extra-grammatical explanations for the observed
acceptability patterns. In the next experiment, the final one on was-f?ur split , we
will scrutinize the role object scrambling plays with respect to subject subextraction,
a state of affairs M?uller (2010) refers to as Melting.
97
3.5 Experiment 4 - Melting
3.5.1 Introduction
Recall from section 3.1.3.2 that M?uller (2010) reports the following contrast:
(37) a. *Was
what
haben
have
denn
PRT
f?ur
for
Leute
people
den
the
Fritz
Fritz
getroffen?
met
?What kind of people met Fritz??
b. Was
what
haben
have
den
the
Fritzi
Fritz
denn
PRT
f?ur
for
Leute
people
ti getroffen?
met
?What kind of people met Fritz??
[M?uller 2010, p.61]
M?uller claims that extraction out of the external subject is only possible if the object
is scrambled across it. This would follow from, indeed be a prediction of, his theory
and his theory only (see section 3.1.3.2 and M?uller?s paper for the details of his
account). This datum is intriguing but some scepticism is warranted whether it in
fact holds as reported. There are a number of issues.
First, M?uller follows the tradition of imposing a binary distinction onto a
continuum. Neither (37-a) or (37-b) are perfect. Is there a difference in the direction
he suggests? It is quite possible but the difference in acceptability is certainly not
significant enough to warrant putting an asterisk on one and nothing at all on the
other. M?uller points out that ?the data have been checked with a number of native
speakers? but does not explicate how many speakers ?a number of speakers? is, what
kind of native speakers he consulted (linguistic colleagues or naive informants), what
kind of task was employed etc. As discussed at some length in section 2.2, this is
a fine practice for clear-cut cases but methodologically inadequate for controversial
data such as (37).30
30Of course, it could be contested as to what constitutes controversial data. Since people?s
opinions on this point are also likely to differ, I suggest erring on the side of caution and consid-
98
Second, all things are not equal in (37). (37-b) differs from (37-a) in that it
has a marked word order, which changes the information structure. So even if the
contrast holds as reported, it is not clear that it is due to a grammatical difference.
In this experiment we will address the status of the data, as given by M?uller , and
we will see that it does not hold as reported. We will also see, however, that it is
tricky to control for all extra-grammatical factors. Let me explain why.
We know that (i) marked order affects acceptability, that is SUB > OBJ is
preferred to OBJ > SUB. We have also known since Behagel (1923) that, ceteris
paribus, (ii) the order definite > indefinite is preferred as compared to indefinite >
definite. Finally, we saw in in section 3.3 that (iii) was-f?ur split does not fare well
when another indefinite argument is present. If we apply the facts in (i) - (iii) to
(37-b), we see that according to (i) its acceptability should be decreased compared
to (37-a) while it should be increased according to (ii). If we try to control for these
conflicting tensions by making the object indefinite, we run into the problem posed
by (iii).
This unholy trinity makes it very challenging toproduce a perfect set ofstimuli.
For this reason we only look at the kind of examples M?uller discusses, and we find
that his data does not hold as reported. It does not, however, fully disentangle
the tension between (i) and (ii). Ultimately, a full-fledged empirical investigation of
Melting will have to take into consideration context effects, i.e. presents the stimuli
providing a specific context. This is left open for future work.
3.5.2 Methodology and Design
37 native speakers of German (by self-assessment) with no prior training in linguis-
tics (a different set of speakers from Experiment 3), were asked to rate sentences
ering every piece of data that is not crystal-clear and beyond doubt to any speaker imaginable as
controversial.
99
on a 7-point Likert scale presented to them in an online questionnaire. The data of
all 37 participants was used for data analysis. The same methodological protocol
outlined in 2.2.2 and in the previous experiments was followed.
This experiment compared extraction out of external subjects with and with-
out the scrambled object as well as extraction out of the object. A subject and an
object question without was-f?ur split were added as a control. Only animate NPs
were used. This gives us the following set of stimuli:
(38) a. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
ein
a
K?afer
beetle
gestern
yesterday
am
on.the
Nachmittag
afternoon
den
the
Beamten
clerk
gebissen
bitten
hat.
has [sub, -ext]
?I ask myself what kind of beetle bit the clerk yesterday afternoon.?
b. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
einen
a
Beamten
clerk
der
the
K?afer
beetle
gestern
yesterday
am
on.the
Nachmittag
afternoon
gebissen
bitten
hat.
has [obj, -ext]
?I ask myself what kind of clerk the beetle bit yesterday afternoon.?
c. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
K?afer
beetle
den
the
Beamten
clerk
gebissen
bitten
hat.
has
[-melting]
?I ask myself what kind of beetle bit the clerk yesterday afternoon.?
d. Ich
I
frage
ask
mich,
myself
was
what
den
the
Beamten
clerk
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
K?afer
beetle
gebissen
bitten
hat.
has
[+melting]
?I ask myself what kind of beetle bit the clerk yesterday afternoon.?
e. Ich
I
frage
ask
mich,
myself
was
what
der
the
K?afer
beetle
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
einen
a
Beamten
clerk
gebissen
bitten
hat.
has
[object, +ext]
?I ask myself what kind of clerk the beetle bit yesterday afternoon.?
100
3.5.3 Results
Unsurprisingly, we find an asymmetry between subject and object question without
was-f?ur split . The object condition is signficantly degraded compared to the subject
condition(5.12 vs. 5.93, t(1,80)=3.04, p=.0014). This is very likely to be an effect
of marked word order O > S (cf. Lenerz 1977, M?uller 1999).
Turning to the extraction conditions, our results show that scrambling of the
object does not improve the subextraction out of subjects (3.57 vs. 3.88), as illus-
trated in Figure 3.9. The difference is not significant: (t<1). Furthermore, both
subject extraction conditions are significantly less acceptable than the extraction out
of objects control: (t(1,80)=1.74, p=.041) for objects vs. subject (without object
scrambling) and (t(1,80)=2.6, p<.004) objects vs. subject (with object scrambling).
This is the exact reverse of what we found for the -extraction condition, (38-a) vs.
(38-b), suggesting that the cost of subextraction out of subjects outweighs the cost
incured by the marked word order effect.
Looking at the individual distribution of the data reveals that only 3 out of the
27 speakers (11%) tested show a significant preference for the subject subextraction
with scrambling compared to the one without scrambling.
3.5.4 Discussion
The experiment shows that M?uller?s judgments does not seem to reflect a general
pattern. Only 11% of the speakers tested share his intuition that object scrambling
improves the acceptability of subject subextraction. Surely, the data of these 3
speakers is relevant and their behavior is predicted by M?uller?s theory. But what
about the remaining 89 %?
This data needs to be handled with care. As discussed above, a number of
factors could affect the acceptability pattern. While we controlled for animacy and
definiteness, it is well known that O-S word order in German results in a marked
101
gid17gid13gid16gid16gid1
gid18gid13gid16gid16gid1
gid19gid13gid16gid16gid1
gid20gid13gid16gid16gid1
gid21gid13gid16gid16gid1
gid22gid13gid16gid16gid1
gid23gid13gid16gid16gid1
gid8gid10gid2gid12gid1gid14gid3gid11gid9gid1 gid7gid2gid4gid12gid1gid14gid3gid11gid9gid1 gid8gid10gid2gid12gid1gid24gid3gid11gid9gid1 gid8gid10gid2gid12gid1gid24gid3gid11gid9gid12gid1
gid24gid6gid3gid5gid9gid1
gid7gid2gid4gid12gid1gid24gid3gid11gid9gid1
Figure 3.9: Experiment 4: Results
information structure, which notoriously reduces the acceptability when judged in
out-of-the-blue contexts. We see this in action looking at the contrast between
(38-a) vs. (38-b). The object question is degraded compared to the subject question,
presumably because of the marked word order it creates.
To counterbalance these effects the target items would need to be embedded in
appropriate contexts that make the marked word order more natural. Unfortunately,
that is easier said than done for various reasons. First, it is a challenge to the creative
capacity of the researcher to construct appropriate and similarly felicitous scenarios
for dozens of sentences (see for instance Fanselow et. al.?s (2008) struggle to create
appropriate contexts to control for the O-S word order created by superiority in
German).
But even the most skillful experimenter cannot avoided the problem created
by H?ohle?s (1982) insight that the marked word order is always a subset of the
unmarked word order with respect to its pragmatic felicity. In other words, the
102
unmarked order is felicitous in every marked context, while marked word orders are
not felicitous in an unmarked context. To illustrate consider the following example:
(39) a. Ein
a.NOM
Mann
man
hat
has
das
the.ACC
Schaf
sheep
gestreichelt.
petted
?A man petted the sheep.?
b. Das
the.ACC
Schaf
sheep
hat
has
ein
a.NOM
Mann
man
gestreichelt.
petted
?As for the sheep, a man petted it.?
(39-a) but not (39-b) is a felicitous utterance in an out-of-the-blue context, e.g. as
an answer to a question like What happened?. (39-b) is appropriate in a context
where the sheep is old information, e.g. as an answer to the question What happened
to the sheep?. (39-a), however, covers all the contexts (39-b) does and as such would
also be appropriate in a context where the sheep is old information, albeit with a
different prosodic stress. So even if (39-b) were appropriate in a certain context, a
speaker might still prefer (39-a), as S-O sequences are considerably more frequent
than O-S sequences.
The bottom line is that M?uller ?s melting effects, as reported in his paper, are
not confirmed by studies of a larger group. This does not necessarily imply that
scrambling of the objects does not affect subextraction out of subjects. There might
still be some reality to this intuition and we leave it up to future research to test
the acceptability under improved methodological conditions (e.g. adding context)
and to provide further discussion about how much of the differences in acceptability
found have grammatical and how much have an extra-grammatical source.
103
3.6 Accounting for the data
3.6.1 The LCA in SOV languages
In the following section we would like to give an account of the experimental find-
ings reported in this chapter in terms of Uriagereka?s (1999) Multiple Spell-Out
(MSO) account, which we already summarized in section 2.1.2.1. Before we move
on, however, let us step back and look at how the LCA handles an SOV language
like German and what the implications for an account in terms of MSO are.
Kayne (1994) departs from the assumption that there is a head parameter
in the grammar which differentiates SVO languages like English or French from
SOV languages like German or Japanese. He tries to derive a universal Spec-Head-
Complement order from the LCA, with head-final sequences being the result of overt
raising of the complement to a position asymmetrically c-commanding (a-CCing)
the head. The assumption that SVO is the only possible base order has received
strong opposition, unsurprisingly mainly by linguists speaking SOV languages, on
both theoretical and empirical grounds (see Rohrbacher (1994) and Fukui & Takano
(1998) and the references cited therein).
On the empirical side, Kayne?s proposal seems to be making the incorrect
prediction that SOV languages should not allow extraction out of objects as a result
of freezing, i.e. if the object position is derived, extraction would originate from a
moved domain. This prediction is quite obviously wrong for languages like German,
Dutch or Japanese.31. To take a concrete case, consider the example in (40). If
the SOV order in (40-a) were derived from an underlying SVO order as indicated
in (40-b), we would expect from our discussion in section 2.1.1.3 that was-f?ur split
should be blocked as a result of freezing. The extraction out of the object position,
31Kayne (p.c.), addressing this issue, points out that the subextraction could take place from
the base position while the SOV word order is derived by subsequent remnant movement. This is
certainly possible but not without complications. For discussion and problems with this proposal
see Abels & Neeleman (2009).
104
however, is highly acceptable, as confirmed by the data presented in this chapter,
which strongly indicates that the object is in an in-situ position and SOV is the
base word order, as indicated in (40-c).
(40) a. Ich
I
wei?
know
nicht,
not
was
what
die
the
Ameise
ant
f?ur
for
einen
a.ACC
Beamten
clerk
gebissen
bitten
hat?
has
?I don?t know what kind of ant bit the clerk??
b. Ich wei? nicht, wasi die Ameise [ti f?ur einen Beamten]j gebissen hat tj?
c. Ich wei? nicht, wasi die Ameise [ ti f?ur einen Beamten] gebissen hat ?
But even on the conceptual side, as pointed out by Rohrbacher (1994), a univer-
sal SVO order does not follow directly from the LCA, contrary to Kayne?s claim.
Concretely, the following statement, found in Kayne (1994:35), is not a direct con-
sequence of the LCA:
?It will always be the case, in any phrase marker, that specifier (S) and
complement (C) are on opposite sides of the head (H) [...] of the six
permutations of H, SandC onlytwo arepermitted by thetheory, namely,
S-H-C and C-H-S. The other four (S-C-H, C-S-H, H-S-C, H-C-S) are
all excluded by the requirement that specifier and complement be on
opposite sides of the head. [my emphasis, JJ]?
The fact that this statement does not follow from the LCA is illustrated in the SVO
and SOV trees in (41):
(41)
a. VP
NP1
N1
n1
VP
V
v
NP2
N2
n2
b. VP
NP1
N1
n1
VP
NP2
N2
n2
V
v
105
In both trees a-CC establishes the same hierarchical relations between the non-
terminals (see Kayne 1994: Chapters 1 -4 for the details of the theory), i.e. the LCA
yields the following set of D-relations between non-terminals, (42-a), and hence the
following set of non-terminal-to-terminal dominance relations d, (42-b):
(42) a. D(A) = {<NP1,VP>, < NP1,V>, <NP1,NP2>, <NP1,N2>, <V,N2>}
b. d(A) = {<n1,v>, <v,n2>, <n1,n2>}
Rohrbacher points out that Kayne?s claim according to which the statements <n1,v>
and <v,n2> together imply that n1 and n2 have to be on opposite sides of v only
if ordering relations of the type <x,y> are sequential in nature and not structural.
Concretely, the pair <n1,v> in (42-b) ?translates into n1 is dominated by a non-
terminal which asymmetrically c-commands a non-terminal dominating v, but not
into n1 spatially [or temporally] precedes [or follows] v ? (Rohrbacher 1994:14). He
illustrates this with the following analogy, which I slightly personalized: in the 1928
Amsterdam Olympics Weightlifting competition the Men?s Individual Featherweight
event finished with Franz Andrysek of Austria, the author?s granduncle (!), winning
the gold medal, Pierino Gabetti of Italy winning silver and Hans W?olpert of Ger-
many winning bronze. This situation can be described in set-theoretical terms:
(43) { <Andrysek, Gabetti >, <Andrysek,W?olpert>, <Gabetti,W?olpert> }
These ordered pairs only describe the relation ?x is a better weightlifter than y?,
or ?x?s performance was better than y?s? etc. There is no implication, however, as
to any sequential or temporal ordering of these three individuals. In fact, the set
in (43) can be represented as in Figure 3.10, where the gold medal winner is in
the middle between silver and bronze (which is common practice at the Olympics).
There is no reason to rule out the possibility that languages differ in how they map
106
d(A) sets into temporal sequences. This, in essence, amounts to saying that there
exists a head parameter.32
Figure 3.10: The Olympic podium
3.6.1.1 The LCA as last resort
While Rohrbacher?s point is well-taken, it is now less clear how the mapping from
ordered sets in the d(A) to actual temporal sequences should work. The only two
straightforward options are to (i) follow Kayne and assume that a-CC always maps
into linear precedence (yielding SVO) or (ii) that it maps into subsequence (yielding
OVS, which is typologically implausible). The mapping algorithm from the d(A)
<S, V>, <S, O>, <V, O> to the actual sequence SOV has to be more complicated.
The system needs a way of keeping track of what the specifiers are, because they
always have to precede the heads, and what the complements are, because they
always have to follow the head. This algorithm, however, applies to terminals which
do not contain this information. In addition, it is not clear how these different
mappings should be formulated in terms of parameters, short of merely restating
the facts, i.e. in SVO languages the algorithm maps into SVO, in SOV it maps into
SOV etc. We conclude that neither Kayne?s way of deriving SOV orders in terms
of movement nor modifying the mapping algorithm from d(A) sets to temporal
32This is essentially the position taken by Richards (2004), Saito & Fukui (1998); see also Fukui
& Takano (1998), who assume SOV orders to be more basic and derive SVO structures through
head movement.
107
sequences is satisfactory. The former runs into a number of empirical problems and
lacks independent motivation while the latter is undesirable from conceptual and
from a learnability point of view.
I will suggest a different solution:
(44) Linear ordering between a head and its first-merged complement is lexically
determined by the head. The LCA applies elsewhere.
(44) essentially reintroduces the head-parameter.33 The idea is that a lexcial head
determines the linear ordering with respect to its first-merged complement. All
other ordering relations are handled by the LCA.
The LCA has always had a troubled relationship with heads and complements.
For simplex complements Kayne created the asymmetry by employing unary branch-
ing for these complements as in (41). He then stipulated the distinction between
categories and segments and redefined c-command to only apply to categories but
not to segments (see Kayne 1994:16ff. for the details).
Furthermore, many languages are not consistently Spec-Head-Comp or Spec-
Comp-Head but show mixed patterns depending on the categories. It is well known
that in German VPs are head-final while PPs are head-initial, as illustrated in the
following example:
(45) Es
it
scheint,
seems
dass
that
der
the.NOM
Hermes
Hermes
den
the.ACC
Kurt
Kurt
in
in
der
the
Scheune
shed
versteckt.
hides
?It seems that Hermes is hiding Kurt in the shed.?
The VP shows the order [[den Kurt] versteckt V] (Comp-Head), while the PP has
the order [P in [der Scheune]] (Head-Comp). To make things worse, even within the
33Similar proposals are made by Saito & Fukui (1998) and Richards (2004).
108
same category there is lexical variation in the ordering of head and complement (the
head is marked in boldface).
(46) a. in
in
den
the.ACC
Flu?
river
?in the.ACC river?
b. *den
the.ACC
Flu?
river
in
in
?in the.ACC river?
c. den
the.ACC
Flu?
river
entlang
along
?along the river?
d. *entlang
along
den
the.ACC
Flu?
river
?along the river?
This leads us to the conclusion that the ordering relation between heads and com-
plements is an idiosyncratic lexical property of the head and does not fall under the
LCA. Only when the ordering relation between two non-heads needs to be deter-
mined does the system call upon the LCA as a last resort mechanism. This allows
us to retain Kayne?s insight that specifiers are always left and that movement is (at
least largely) to the left34, while it gives us the flexibility in the head-complement
domain needed to capture the inter -and intralinguistic variation we find.35
Taking out the LCA from the domain of heads and their complements allows
us to maintain MSO for SOV languages. If SOV word orders are not the results of
movement, we do not predict any freezing effects for extraction out of complements,
34There are a number of cases of movement to the right, e.g. extraposition, heavy NP shift,
rightward focus etc. Whether this is genuine syntactic movement or movement at PF is still
a matter of debate. See Drummond (2009) for a recent of proposal of rightward movement as
genuine synactic movement and section 4.3 for further discussion of extraposition. How syntactic
movement to the right can be reconciled with the LCA is unclear.
35One potentially problematic case arises when a simplex specifier such as a proper name is
merged. In strict BPS terms this specifier would also be a head and the LCA would not apply. It
is unclear, however, whether simplex specifiers really exist (see Burge 1973 and Longobardi 1994
for arguments that even proper names are internally complex). In addition, BPS is often assumed
to be at work in narrow syntax only, while more specific information needs to be available at the
interfaces (Chomsky 1995). If so, PF would recognize simplex specifiers as XPs (rather than heads)
and the LCA can apply.
109
as is shown in (40). Even though linearization in terms of a-CC relations is at the
core of MSO, it does not seem to be necessary to follow Kayne in all details. Recall
that the core insight of the MSO theory is that phrase markers assembled in a sep-
arate workspace need to be spelled-out to the interfaces before they can be merged
with the main spine. This requirement is forced upon the system by PF-linearization
requirements defined in terms of a-CC. However, nothing in principle prevents us
from departing from Kayne?s assumption that directionality is universally fixed as
S-H-C and that all other sequences are derived by movement. We can instead retain
the insight of MSO, while acknowledging the conceptual and empirical advantages
of a head parameter. Hence, for the purpose of this paper we will assume that both
SVO as well as SOV sequences are base-generated.
3.6.2 A MSO Account of was-f?ur split
We will now sketch how the was-f?ur split data is accounted for. The cases of
extraction out of non-finite clauses in German discussed in the following chapter
work in a parallel way.
Let us turn to Was-f?ur split of the kind explored in this chapter. Consider
the VP in a case of extraction out of VP-internal subjects as in (14-c) (repeated in
(47)) with a structure as in (48).36
(47) Was
what
hat
has
denn
indeed
f?ur
for
eine
an
Ameise
ant
den
the
Beamten
clerk
gebissen?
bitten
?What kind of ant bit the clerk??
36We follow Leu (2008) in assuming that Was-f?ur phrases are whPs and that was subextracts
out of them. We will remain agnostic to the details of the structure.
110
(48) VP
whP
was f?ur eine Ameise
V?
DP
den Beamten
V0
gebissen
How did this phrase marker come about? As we saw in section 2.1.2.1, it is in
inevitability of a strictly bottom-up derivational system that complex specifiers need
to be constructed in a separate workspace before they can be merged with the main
spine. It follows that the two phrase markers in (49) are assembled in parallel.
(49)
whP
was PP
f?ur eine Ameise
VP
V?
DP
den Beamten
V0
gebissen
Note that we are assuming, contrary to Kayne (1994), that SOV structures are
base-generated (see the discussion above).
Recall that according to (44) only the ordering between the head ?gebissen? and
its complement is determined lexically. The total ordering of all other terminals is
established through the LCA. In order for the LCA to yield a complete set of ordered
pairs of precedence relations between the terminal elements, the specifier is spelled
out independently and then merged with the main spine, yielding (50).
111
(50) VP
DP<was,f?ur,eine,Ameise> V?
DP
den Beamten
V0
gebissen
The elements inside the whP have been shipped off to the interfaces and can no
longer be affected by syntactic operations. Now T andC aremerged. The wh-feature
on C cannot probe for and attract was, since the latter is no longer accessible for
syntax. As a result, the feature on C remains unchecked and the derivation crashes.
How are subject questions such as (51) derived, where the entire was-f?ur phrase
moves?
(51) Was
what
f?ur
for
eine
an
Ameise
ant
hat
has
denn
indeed
den
the
Beamten
clerk
gebissen?
bitten
?What kind of clerk did the ant bite??
Note that even though the substructure <was,f?ur,eine,Ameise> has been spelled-
out, the entire DP is still accessible. We to assume that the wh-feature on the ?was?
percolates to the DP prior to SO and is still accessible even after ?was? is spelled-out.
As such, the wh-feature on C can be checked by movement of the full DP.
The MSO proposal correctly predicts Was-f?ur split out of (unmoved) objects
to be grammatical. For an example like (14-e), repeated as (52), we have a structure
such as (53).
(52) Was
what
hat
has
denn
indeed
die
the
Ameise
ant
f?ur
for
einen
a
Beamten
clerk
gebissen?
bitten
?What kind of clerk did the ant bite??
112
(53) VP
DP
der Beamte
V?
whP
was f?ur eine Ameise
V0
gebissen
The object can be assembled in the main workspace. First, the whP is put together
and as a result does not need to be spelled-out until the final step of the derivation.
After merger of T and C, the wh-feature can probe down the main spine and finds
was, which is copied and moved to SpecCP. The derivation converges.
Note that the same logic used to rule out subextraction out of subjects applies
to subextraction out of indirect objects, as long as they are left branches. Combing
the MSO with Baker?s (1988) and Larson?s (1988) analyses of indirect objects as
complements of V, would predict them to allow extraction. We saw in Experiment
2 that this is likely to be empirically incorrect, which is a strong reason to doubt
Baker?s and Larson?s take on the position of indirect objects.
3.7 ATB was-f?ur split
In this section we would like to present a phenomenon that, to our knowledge,
has not been discussed in the German literature so far. We would like to call
this construction Across-the-Board (ATB)was-f?ur split . Consider the following
examples:
113
(54) a. Was
was
hat
hat
denn
PRT
f?ur
for
eine
a
Prinzessin
princess
f?ur
for
einen
a
Frosch
frog
gek?usst?
kissed
?What kind of princess kissed what kind of frog??
b. Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
M?adchen
girl
f?ur
for
ein
a
Mann
man
vorgestellt?
introduced
?What kind of man was what kind of girl introduced to??
Note that this construction is different from wh-exclamatives. Consider these ex-
amples from Austrian German:
(55) a. Was
what
hat
has
die
the
Weink?onigin
wine.queen
nur
only
f?ur
for
einen
a
Ungustl
scumbag
geheiratet!
married
?How could the wine queen marry such a scumbag!?
b. Was
what
haben
have
nur
only
f?ur
for
Politiker
politicians
f?ur
for
an
a.ACC
Bledsinn
nonsense
zamgredet!
talked
?How could these sorts of politicians talk this sort of nonsense!?
The utterance in (55) does not request information, but is a statement of amazement
or disbelief. Holding wh-exclamatives and was-f?ur split apart becomes particularly
important in light of the discussion of these phenomena in Dutch brought to my
awareness by Norbert Corver (p.c.). Corver reports the following contrast for Dutch:
(56) a. Wat
what
hebben
have
er
there
?n
a
mensen
people
?n
a
boeken
books
gekocht!
bought
[Dutch]
?So many people bought so many books!?
b. *Wat
what
hebben
have
er
there
voor
for
n?
a
mensen
people
voor
for
?n
a
boeken
books
gekocht?
bought
?What kind of people bought what kind of books??
In Dutch a single wat cannot be associated with two voor remnants, as shown in
(56-b). This contrasts with the wat-exclamative in (56-a). Corver (1991) takes this
contrast as an argument that the wat-exclamative construction, unlike wat voor-
split, does not involve wh-extraction. While both constructions seem remarkably
114
similar in German and Dutch, German was-f?ur split allows one fronted was to be
associated with two f?ur remnants, as shown in (54). These examples could not
be parsed as wh-exclamatives since the particle denn is only licensed in a question
context. We conclude that German allows true ATB was-f?ur split . It is beyond the
scope of this paper to explain why Dutch and German pattern differently in this
respect.
As the English translations of (54) suggest, we are truly asking multiple ques-
tions about pairs of individuals. Possible answers for (54-a) could be: A pretty
princess kissed an ugly frog, A tall princess kissed a green frog, A purely dressed
princess kissed a happy frog, etc. Note that the qualities we are quantifying over do
not have to be of the same type, i.e. you can talk about the height of individual A
and the color of individual B etc.
Another interesting property of ATB was-f?ur split is that for some speakers it
can ameliorate island violations very similar to Parasitic Gap (PG) constructions,
i.e. a good gap licenses a bad gap:37
(57) Extraction out of subjects repaired by ATB-movement
a. ?*Was
was
hat
hat
denn
PRT
f?ur
for
eine
a
Prinzessin
princess
einen
a
Frosch
frog
gek?usst?
kissed
?What kind of princess kissed a frog??
b. Was
was
hat
hat
denn
PRT
f?ur
for
eine
a
Prinzessin
princess
f?ur
for
einen
a
Frosch
frog
gek?usst?
kissed
?What kind of princess kissed what kind of frog??
37In a sense, PG-was-f?ur split might be the more appropriate term.
115
(58) Extraction out of IOs repaired by ATB-movement
a. *Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
M?adchen
girl
ein
a
Mann
man
vorgestellt?
introduced
?What kind of man was what kind of girl introduced to??
b. Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
M?adchen
girl
f?ur
for
ein
a
Mann
man
vorgestellt?
introduced
?What kind of man was what kind of girl introduced to??
What do we mean by some speakers? It this were a regular paper in theoretical
syntax we would leave it at that and quote the examples with our judgments. Since
we have spent a considerable amount of space discussing gradience in grammar and
the need to gather data from a statistically significant group of speakers and not
just a few individuals, we need to be a little more precise. When testing these con-
structions using the usual experimental methodology (n=23), 52% of the speakers
patterned in the way reported in (57) and (58), i.e. they preferred ATB was-f?ur
split to simple was-f?ur split out of islands. 21% did not show this pattern and with
26% you could not tell since for this group was-f?ur split only seemed a marginal
construction to begin with. For the last group we only found floor effects and po-
tential differences were washed out. Interestingly, when informally asking German
speaking linguists about this construction, they all shared the judgments in (58) and
(57). This suggests that ATB was-f?ur split is a stretch for many non-linguistically
trained speakers. As we saw above, it involves a pair-list question about two sets
of quality types. It is not easy to construct a plausible situation where this would
be an appropriate question, a factor that often strongly influences speakers? judg-
ments. Additionally, was-f?ur split is a relatively informal construction and, despite
our efforts to clarify that we were not interested in prescriptive judgments, it seems
highly likely that some speakers were still affected by a prescriptivist bias.38
38There is no prescriptive rule about was-f?ur split in traditional German grammar books. How-
ever, it still feels like the kind of construction a high school German teacher would cross out as
too informal and inappropriate for an essay.
116
That said, we believe that ATB was-f?ur split is a genuine phenomenon of
German and that the PG-like effects robustly hold. In the next section we will see
how a sideward movement account of ATB movement (Nunes 2001, Hornstein &
Nunes 2002) of ATB movement can provide an explanation of these facts.
3.7.1 A sideward movement account of ATB was-f?ur split
The copy theory of movement views syntactic movement as a complex operation
that can be decomposed into Copy and Merge. In (59), Y is copied, as indicated
by the angled brackets, and merged higher up in the same tree. In (60), Sideward
Movement (SM) applies. Y is copied and sideward moved into a separate phrase
marker, which has been constructed at the same time.
(59) ZP
Y XP
X <Y>
(60)
XP
X <Y>
KP
Y AP
A B
SM has been used to account for PG and ATB movement by Nunes (2001), Nunes
& Uriagereka (2000) and Hornstein & Nunes (2002). We will try to show that these
analyses nicely carry over to provide an account for ATB was-f?ur split and also
explain its property of island repair.
Consider the derivation of (57-b), repeated as (61-a), with a slightly simplified
numeration such as (61-b) (For expository purposes the English glosses are used for
the derivation. Of course, we are in fact deriving the German sentence.)
117
(61) a. Was
was
hat
hat
denn
PRT
f?ur
for
eine
a
Prinzessin
princess
f?ur
for
einen
a
Frosch
frog
gek?usst?
kissed
?What kind of princess kissed what kind of frog??
b. N={T, C, what, has, PRT, for, a, princess, for, a, frog, kissed}
As a first step the object is constructed and merged with the verb to form the main
spine. Simultaneously, the specifier is assembled in a separate workspace, as shown
in (62).
(62)
whP
PP
for DP
a princess
VP
VP
whP
what for a frog
V0
kissed
Now, we are left with the following elements in the numeration:
(63) N={T, C, what, has, for, a, princess, for, a, frog, kissed}
In order for the derivation to converge, what needs to be merged in the specifier of
whP. German does not allow a for-PP to feature as thematic agent of a sentence.
Since there is no what left in the numeration to be merged, the only way for the
derivation to converge is to form a copy of what in the object position and sideward
move it to the specifier of whP, as illustrated in (64):
118
(64)
whP
<what> PP
for DP
a princess
VP
VP
whP
<what> for a frog
V0
kissed
The next step is to spell-out the specifier before merging it with the main spine.
Recall that this step is necessary in order for the linearization algorithm to yield a
total set of precedence relations (see section 2.1.2.1 and for the details). This results
in the structure in (65):
(65) VP
whP<what,for,a,princess> VP
whP
what for a frog
V0
kissed
In the next steps T and C are merged. C has a Q-feature that needs to be checked.
Probing down the tree, the only wh-element available is the what in the object
position. The what in the subject position is no longer accessible since the entire
subject whP has already been spelled-out. Norbert Hornstein (p.c.) notes that the
entire whP could be a target and would thus block the attraction of the ?was? in
the object position. We have argued in section 3.6.2 that the wh-feature of ?was?
can percolate to whP to allow for a derivation where the entire was-f?ur phrase
moves to SpecCP. The only way of reconciling this situation is by assuming that the
percolation of the wh-feature to whP is optional. If it does, the whP is a target for
the probing feature on C and moves the SpecCP. This would correctly derive the
119
grammatical example in (66):
(66) Was f?ur eine Prinzessin hat denn was f?ur einen Frosch gek?usst?
was for a princess has PRT was for a frog kissed
?What kind of princess kissed what kind of frog??
If the wh-feature does not percolate, the What is copied and merged in SpecCP.
Nunes? (2004) chain reduction algorithm makes sure that only the highest copy is
pronounced while all other copies are deleted. As a result, the copy of what in the
subject position and in the object position are deleted, as illustrated in (67).
(67) CP
what C?
C[+Q] TP
T0
has
VP
whP<what,for,a,princess> VP
whP
what for a frog
V0
kissed
It is easy to see how the convergence of the derivation hinges on the copy of what in
the object position. If it were not available, the feature on C could not be checked.
This is exactly the scenario in non ATB was-f?ur split contexts, when extraction out
of island takes places, hence the ungrammaticality of (57-a) and (58-a). The SM
120
step salvages the derivation.
3.8 Conclusion
This chapter summarized the first set of experiments on subject island effects in
German. Data from a number of studies on was-f?ur split converge on the conclusion
that extraction out of subjects is degraded compared to extraction out of objects and
that these subject island effects are independent of freezing effects. We provided
a number of arguments that the patterns of acceptability we find are in fact a
reflection of anunderlying grammatical difference. Factorslike informationstructure
and parsing certainly play a role in acceptability studies but the body of data we
gathered seems to convincingly rule out the possibility that all gradience solely stems
from these extra-grammatical factors.
Theoretically, we have argued the MSO provides a straightforward account
for the asymmetries we found, under a slight modification of the LCA. Essentially,
complements are the preferred to domain for subextraction. We emphasized the
point that a grammatical constraint always leads to a decrease in acceptability but
does not necessarily lead to full unacceptability. We also noted that MSO predicts
freezing of moved complements but has nothing to say about the additional decrease
in acceptability of extraction out of moved specifiers.
Finally, we discussed ATB was-f?ur split , a phenomenon not discussed so far
in the German literature. We saw that ATB movement of was can salvage island
violations in a way similar to other instances of ATB-movement and parasitic gap
constructions. We provided an analysis in terms of Sidewards Movement.
In the following chapter we will turn to a detailed experimental investigation
of subextraction out of sentential domains in German. We will conclude that the
results we found provide further evidence for the existence of the CED in German.
121
Chapter 4
Extraction out of non-finite clauses in German
4.1 Introduction
In this chapter we will investigate wh-extraction out of non-finite clauses in German.
We will first summarize the discussion in the theoretical literature and conclude that
more careful experimental elicitation of the facts is warranted. In particular, we will
not only look at extraction out of sentential subjects in isolation, but compare them
to extraction out of sentential objects as well. We will arrive at the same conclu-
sion as in our was-f?ur split experiments: Extraction out of subjects is degraded as
compared to extraction out of objects.
4.2 Background
The question of whether German allows wh-extraction out of sentential subjects
has occupied syntacticans at least since the early eighties. The first more detailed
discussion of the facts can be found in Haider (1983), who argues that German does
not obey the subject condition.
122
(1) a. Welchen
which.ACC
Polizisten
policeman
glaubst
think
du
you
w?urde
would
damit
with.it
zu
to
t?auschen
trick
selbst
even
diesem
this.DAT
Gauner
crook
schwerfallen?
fall.hard
?(*)Which policeman do you think that tricking would be hard even for
this crook??
b. Welches
which
Buch
book
meinen
think
Sie
you
ist
is
heuer
this.year
zu
to
praemieren
award
beschlossen
decided
worden?
been
?(?*)Which book do you think (it) was decided to give an award to this
year??
[Haider (1983), p. 94-95, ex. (17) and (18)]
Working in Kayne?s Connectedness framework, Haider attributes the difference be-
tween English and Dutch on the one hand, where subjects are islands for extraction,
and German, on the other hand, to be a consequence of the obligatory subject rais-
ing in English and Dutch to an ungoverned position. The subject in German stays
within the g-projection of V, and the trace of the subextraction satisfies the ECP.
Note that Haider?s position is descriptively identical to Stepanov?s point of view
some 20 years later: there is nothing illicit about extraction out of subjects per se
but subjects become opaque if they are moved as a result of freezing effects.
Sternefeld (1985) argues that Haider?s examples are not genuine cases of ex-
traction out of subjects because they involve unaccusative and passive predicates,
where the structural subject is an underlying object. If this is controlled for, the
subject condition also holds in German:
123
(2) a. *Wem
who.DAT
geh?ort
befit
sich
self
nicht,
not
Briefe
letters
nicht
not
zu
to
beantworten.
answer?
(*)Who is not answering letters to inappropriate?
b. *Was
what
bezeugt
attest
zu
to
finden
find
guten
good
Orientierungssinn?
sense.of.orientation
?(*)What does to find indicate a good sense of orientation??
[Sternefeld (1985), p. 403, ex. (27)]
Note that Sternefeld?s and Haider?s examples are not minimal pairs in two relevant
respects. First, Sternefeld uses non non-d-linked while Haider uses d-linked wh-
arguments. It has been well established that d-linking significantly improves the
extractability of wh-elements1. Second and more interestingly, it has been pointed
out by Grewendorf (1989) that Haider?s and Sternefeld?s examples differ in as far as
the former uses an auxiliary in the V2 position while the latter uses a main verb.
Grewendorf argues that this is the crucial difference determining the acceptability
of this construction:
(3) a. Wessen
whose
Beispiele
examples
hat
has
zu
to
analysieren
analyze
dich
you
mehr
more
frustriert
frustrated
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
?(*?) Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
b. *Wessen
whose
Beispiele
examples
frustrierte
frustrated
dich
you
zu
to
analysieren
analyze
mehr
more
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
?(*?) Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
[Grewendorf (1989]
In (3-a) the V2 position is filled with the auxiliary hat (?has?), and the wh-extraction
1See Pesetsky (1987) and Cinque (1990) for theoretical discussion and Sprouse (2007) for ex-
perimental evidence confirming the ameliorative effect of d-linking.
124
out of the sentential subject is judged acceptable. In (3-b) on the other hand the
V2 position is occupied with the main verb frustrierte (?frustrated?), and the wh-
extraction out of the sentential subject is judged to be illicit. This is quite a peculiar
observation since it is not immediately obvious why the islandhood of the subject
should depend on which element the matrix C is filled with.
Haider (1993:159) acknowledges this contrast but rejects a grammatical expla-
nation of it. He contemplates a parsing account, arguing that the verbal bracketing
in the case of auxiliary-participles - present in (3-a) and absent in (3-b) - somehow
facilitates the processing of the filler-gap dependency. This entails that the pars-
ing difficulty should disappear with separable particle verbs, which also involve a
bracketing from the V2 position to the final position in the clause. Haider claims
that this prediction is indeed borne out. He does not give minimal pairs but judges
extraction out of subjects involving particle verbs with a (?), while assigning ?? to
different examples with non-separable verbs.2
(4) a. ??In
in
welche
which
Akten
files
verwehrte
denied
Einsicht
access
zu
to
nehmen
take
dir
you.DAT
der
the
Richter?
judge
?(?*)In which files did the judge deny you access??
(Haider 1993, 159: ex. (81b), my translation, JJ)
b. (?)Worueber
about.what
fiel
fell
mit
with
dir
you.DAT
zu
to
plaudern
chat
keinem
noone.DAT
ein?
in
?(?*)What did chatting about with you occurred to nobody?
[Haider 1993, 159: ex. (83a), my translation, JJ]
This is another domain where purely introspective gathering of judgments reaches
2Quite oddly, Haider also assigns a (?) to the following example which does not contain a
separable verb. This makes his use of question marks and double question marks even more
mysterious.
i. (?)Wor?uber
about.what
gelang
succeed
mit
with
dir
you.DAT
zu
to
plaudern
chat
keinem?
noone.DAT
?What did chatting about with you nobody succeed??
[Haider 1993, 159: ex. (83c), my translation, JJ]
125
its limits. Haider is careful to qualify his statement stating that the parsing burden
?sollte? (?should?) disappear with particle verbs. Then he goes on to give a contrast
between (?) and ?? between two examples that are not minimal pairs. At no point
does he explicate what (?) or ?? mean in a purely binary view on grammaticality,
which he clearly advocates for in other places (see Haider 1993:159and the discussion
in section 2.2.4). In section 4.5, we will test Haider?s claim and give some content
to the question marks and test whether the parsing burden only should disappear
or actually does.
Haider (1983) also argues that extraction out of extraposed sentential subjects
is only possible if the extraction site is not filled with an expletive, which he argues
to block the formation of a licit chain formation (see Haider (1983) for the details
of the account).
(5) a. *Das
that
ist
is
ein
a
Klavier,
piano
das
that
es
it
mir
me
Spa?
fun
macht
makes
zu
to
spielen.
play
?That?s a piano which it is fun for me to play?
b. Das
that
ist
is
ein
a
Klavier,
piano
das
which
mir
me
Spa?
fun
machte
made
vierh?andig
four-handed
zu
to
spielen.
play
?That?s a piano which it was fun for me to play four-handed?
[Haider (1983), p. 100, ex. (30a) and (31)]
To sum up, it seems as if the islandhood of sentential subjects depends on a number
of factors: (i) whether the subject is in-situ or in a derived position, (ii) whether
the matrix predicate is unaccusative or unergative, (iii) whether the extracted wh-
element is d-linked or not, (iv) whether the V2 position is filled with a main verb
or an auxiliary and (v) whether the sentential subject is extraposed (and whether
the expletive es is present).
In the following sections we will discuss a sequence of acceptability judgments
studies on extraction out of sentential subjects. In 4.3 we will investigate the overall
acceptability of subextraction out of sentential arguments and the role rightward
126
extraposition plays. We will see that extraction out of subjects is significantly de-
graded compared to objects. Extraposition does not influence the extractability for
objects while it marginally does for subjects. Furthermore the presence of the exple-
tive es does not have aneffect on the acceptability of subextraction out of extraposed
subjects, contrary to Haider?s judgments. The experiment in section 4.4 confirms
Grewendorf?s insight that auxiliaries in the V2 position facilitate subextraction as
compared to main verbs. This, however, is not a quirk of subject subextraction
but holds for extraction out of sentential arguments in general. Finally, section 4.5
disconfirms Haider?s speculation that the acceptability of subextraction is tied to
verbal bracketing.
4.3 Experiment 5 - Sentential subjects vs. sentential ob-
jects
4.3.1 Introduction
This experiment is designed to shed light on the overall acceptability of subextrac-
tion out of sentential arguments and the role rightward extraposition plays. We will
again find our familiar and beloved subject/object asymmetry. Extraposition turns
out to have a different effect on subjects compared to objects.
4.3.2 Methodology and design
32 speakers participated in this study, and the same experimental protocol was
followed (again, the reader is referred to section 2.2.2 for the details). The design
had a 2x2x2 structure manipulating the factors Subject/Object, Extraction
and Extraposition. This yields the following 8 conditions:
127
(6) a. Die
The
Diplomarbeit
MA
zu
to
schreiben
write
hat
has
die
the
Studentin
student.FEM
gelangweilt.
bored
?Writing the MA has bored the student.? [sub, -ext]
b. Welche
Which
Arbeit
paper
hat
has
denn
PRT
zu
to
schreiben
write
die
the
Studentin
student.FEM
gelangweilt?
bored
?Which paper has writing bored the student?? [sub, +ext]
c. Die
The
Studentin
student.FEM
hat
has
die
the
Diplomarbeit
MA
zu
to
schreiben
write
vorgehabt.
planned
?The student planned to write the MA.? [obj, -ext]
d. Welche
Which
Arbeit
paper
hat
has
denn
PRT
die
the
Studentin
student.FEM
zu
to
schreiben
write
vorgehabt?
planned
?Which paper has the student planned to write?? [obj, +ext]
e. Es
it
hat
has
die
the
Studentin
student.FEM
gelangweilt
bored
die
the
Diplomarbeit
MA
zu
to
schreiben.
write
?It bored the student to write the MA.? [sub, -ext, +ep]
f. Welche
Which
Arbeit
paper
hat
has
(es)
(it)
denn
PRT
die
the
Studentin
student.FEM
gelangweilt
bored
zu
to
schreiben?
write
?Which paper did it bore the student to write?? [sub, +ext, +ep]
g. Die
The
Studentin
student.FEM
hat
has
vorgehabt
planned
die
the
Diplomarbeit
MA
zu
to
schreiben.
write
?The student planned to write the MA.? [obj, -ext, +ep]
h. Welche
Which
Arbeit
paper
hat
has
denn
PRT
die
the
Studentin
student.FEM
vorgehabt
planned
zu
to
schreiben?
write
?Which paper did the student plan to write?? [obj, +ext, +ep]
Note that we cannot construct ideal minimal pairs in this domain because the pred-
icates that take non-finite sentential subjects and the predicates that take sentential
objects do not overlap. We thus used psych verbs in the subject conditions and
subject control verbs in the object condition. 3 This is a potentially confounding
3Sentential subject predicates included verbs like: langweilen (?bore?), belasten (?strain?), er-
freuen (?delight?), ver?argern (?annoy), etc. and sentential objects predicates verbs like: pla-
nen(?plan?), verlautbaren(?announce?), veranlassen(?bring about?), vorhaben(?intend?). See the ap-
pendix for the full list of stimuli.
Note that it is possible to construct minimal pairs using finite clauses with predicates like
beweisen (?prove?), bestimmen (?determine?) or nahelegen (?suggest?). Extracting out of finite
128
factor to be borne in mind when analyzing the results.
We used d-linked WH-arguments throughout all the conditions to give ex-
traction the best possible chance. As in the was-f?ur split experiments, the particle
denn (?indeed?) marks the VP-boundary (cf. Diesing 1992 and Webelhuth 1989 for
German, Pollock 1989 for French and English).
4.3.3 Predictions
It is a matter of theoretical controversy what the base position of sentential ar-
guments in German is and how extraposition is derived. We will not go into the
intricacies of this debate here. What is relevant for our discussion is the question
whether the non-extraposed sentential arguments in (6-a) to (6-d) have been base-
generated in VP or have come to surface in this position as a result of movement.
If the latter were the case we would expect to find some sort of freezing effects.
Conversely, we may ask whether the extraposed sentential arguments in (6-e) to
(6-h) have been moved to some rightward adjoined position or were base-generated
there. If the former were the case, we would also be interested in whether this
movement took place in narrow syntax or at PF. If it happened in narrow syntax
we would expect freezing effects since the movement would presumably precede the
wh-extraction. If it took place at PF we would not expect any such effects.
All these logical possibilities have been put forth in some form or another in
the literature (see B?uring & Hartmann 1997 for an overview). Even though the
main point of this study is to test whether German shows subject/object asymme-
tries for extractability, we may also shed some light on this discussion. There are
a number of possible outcomes, each of which would be potential evidence for a
specific theoretical position. In the following we will provide a (non-exclusive) list
clauses with such predicates, however, creates some sort of factive islands and is very marginal for
most speakers, which prompted us not to use them as our test items. We did, however, add some
of these cases as fillers, and we also found a subject/object asymmetry, albeit very low on the scale
(1.62 vs. 2.29, p < .001)
129
of some possible scenarios, indicating the theoretical implications (note that ?base?
and ?extraposed? are purely descriptive terms referring to the overt position of the
sentential argument and do not imply any theoretical stance).
(7) WH-Extraction out of sentential arguments - possible scenarios:
a. no sub/obj asymmetry in base or extraposed position
sentential arguments are base-generated in VP, extraction is possible (as
predicted by Stepanov). Extraposition happens after the extraction has
taken place (either late in syntax or post-syntactically) and does not
have an effect on the extractability
b. sub/obj asymmetry in base, no asymmetry when extraposed
Sentential arguments are base-generated in VP and extraction shows
classical CED effects (against Stepanov). The extraposed clauses are
also base-generated (cf. Webelhuth 1989) in complement positions and
thus do not show an asymmetry for extraction.
c. sub/obj asymmetry both in base and extraposed position
Sentential arguments are base-generated in VP and extraction shows
classical CED effects (against Stepanov). Extraposition is rightward
movement to some adjoined position after the extraction has taken
place.4 Hence only the base position determines the permeability of
a domain, extraposition does not change this status. B?uring & Hart-
mann (1997) argue for this position based on extraction, binding and
reconstruction data.
We will see in the next section that our results are compatible with (7-c) but do
not provide conclusive evidence for it. We find a subject/object asymmetry both
4This is essentially the account argued for recently by Drummond (2009). It is unclear, however,
how syntactic movement to the right is reconcilable with the LCA.
130
with and without extraposition. However, subextraction out of subjects is slightly
improved with extraposition, yet still degraded compared to objects.
4.3.4 Results
In both the extraposition and the non-extraposition cases, extraction out of subject
clauses is significantly degraded in comparison to extraction out of objects. There
is a strong interaction effect for the factors Sub/Obj and Extraction. Figure 4.1
and Table 4.1 show the sample means, Table 4.2 shows the results of a three-way
repeated measures ANOVA.
gid20gid17gid19gid19gid1
gid21gid17gid19gid19gid1
gid22gid17gid19gid19gid1
gid23gid17gid19gid19gid1
gid24gid17gid19gid19gid1
gid25gid17gid19gid19gid1
gid26gid17gid19gid19gid1
gid13gid4gid11gid8gid7gid2gid3gid9gid6gid5gid1 gid14gid4gid11gid8gid7gid2gid3gid9gid6gid5gid1
gid3gid15gid5gid9gid8gid6gid14gid1
gid2gid5gid9gid8gid6gid14gid1
gid3gid15gid5gid9gid8gid6gid14gid1gid8gid16gid14gid12gid4gid11gid10gid13gid8gid7gid1
gid2gid5gid9gid8gid6gid14gid1gid8gid16gid14gid12gid4gid11gid10gid13gid8gid7gid1
Figure 4.1: Results
Table 4.1: Means
Non Extraposed Extraposed
Sub, -ext 6.49 6.02
Sub, +ext 3.29 3.98
Obj, -ext 5.14 5.99
Obj, +ext 5.52 6.01
131
Table 4.2: ANOVA
F p
Sub/Obj*Extraction 146.4 ***
Extraction 97.2 ***
Sub/Obj 47.5 ***
Extraposition 11.9 **
Extrapostion*Extraction 2.4 .122
There is both a Sub/Obj main effect as well as an interaction effect for Sub/Obj x
Extraction. In other words, extraction out of a subject incurs a larger cost than
extraction out of the object. The solid black line in Figure 4.1 refers to the non-
extraposed object condition. This line indicates even a slight increase for extraction
as compared to no extraction. Pair-wise comparisons revealed that extraction out
of non-extraposed subject clauses is significantly degraded (3.29) in comparison to
extraction out of non-extraposed objects (5.52) (t(1,92)=8.4, p<.001), while in the
non-extraposed baseline conditions subjects (6.49) are preferred to objects (5.16)
(t(1,92)=6.6, p?.001). This is a result of the fact that many speakers disprefer
phonologically heavy constituents in preverbal object position, cf. (6-c). When the
wh-argument is extracted, enough phonological weight is taken off the constituent
to outweigh the negative effect of extraction, cf. (6-d). This effect evens out in
the extraposition conditions as indicated by the dotted black line. Subject-object
asymmetries for extraction are also found in the extraposed conditions (3.98 vs.
6.01, t(1,92)=7.44, p<.001).
What is the role of extraposition? Both the non-extraposed and the extra-
posed subject conditions (the grey lines in Figure 4.1) indicate a strong decrease in
acceptability for extraction. There is a marginally significant main effect for the fac-
tor Extraposition, i.e. speakers slightly prefer extraposing sentential arguments
across the various conditions, but there is no interaction effect between Extrapo-
sition x Extraction. This suggests that extracting out of extraposed sentential
132
arguments does not change the pattern found in the base cases. The base position
solely determines the status of the extractability.
However, extraposition affects sentential subjects and objects differently. For
this reasons we conducted two ANOVA subanalyses for subjects and objects, mea-
suring the effects of the factors Extraction and Extraposition separately for
the two argument types. The results are summarized in Tables 4.3 and 4.4.
Table 4.3: ANOVA subanalysis for sentential subjects
F p partial ?2
Extraction 236.604 *** .722
Extraposition .382 .538 .004
Extrapostion*Extraction 12.738 ** .123
Table 4.4: ANOVA subanalysis for sentential objects
F p partial ?2
Extraction 1.434 .234 .15
Extraposition 15.782 *** .146
Extrapostion*Extraction 1.214 .273 .013
It turns out that there is an Extraction main effect only for subjects and an Ex-
traposition main effect only for objects. Furthermore, we find an Extrapostion
x Extraction interaction effect only for subjects but not for objects. The partial
?2 square values measure the effect size, i.e. the proportion of the variance that is
accounted for by the independent variable.5 Both the Extraposition main effect
for objects and the Extrapostion*Extraction interaction effect for subjects
are medium sized.
5Following convention, .01 is considered a small effect, .09 a medium effect, and .25 a large
effect (cf. Sprouse 2007, Weskott & Fanselow 2008).
133
4.3.5 Discussion
Our results confirm that there is significant asymmetry with respect to the extract-
bility of wh-elements between sentential subjects and objects. Subextraction out
of sentential subjects is significantly degraded, as expected under the CED. A few
aspects of the data deserve closer discussion.
The object non-extraposed baseline condition, (6-c), is degraded compared to
the extraction counterpart, (6-d). This is peculiar at first glance, for we expect the
creation of a filler-gap depedency to cause a higher processing load which normally
manifests itself in decreased acceptability, a well-established observation in the sen-
tence processing literature (eg. Gibson 1998) and also what we found in the was-f?ur
split experiments. However, there is an interfering factor in the construction at
hand. Speakers quite strongly disprefer non-extraposed sentential objects, which is
likely to be caused by a reluctance to place phonologically heavy constituents in
the Mittelfeld. If the d-linked wh-phrase is moved out of the sentential object and
replaced by a trace, this phonological weight is lifted significantly. This outweighs
whatever cost the filler-gap dependency incurs. This explanation is corroborated by
the fact that the decreased acceptability of the object baseline condition disappears
when the sentential object is extraposed, as in (6-g). Extraposition, however, has no
effect on the acceptability of the extraction. Wh-subextraction is judged as highly
acceptable no matter whether the sentential object is extraposed or not.
The same phonological heaviness effect is not found with sentential subjects.
Here extraposition is even slightly disprefered in the baseline condition, which might
be due to the fact that the sentence starts with an expletive, as in (6-e). Unlike
in the case of objects, extraposition does affect the permeability of sentential sub-
jects. Extraction out of extraposed sentenial subjects, (6-f), is dispropotionally
less degraded than extraction out of the non-extraposed counterpart, (6-b). Both
extractions, however, are still worse than objects.
134
It is not straightforward to determine the implications of the results for the the-
ory of extraposition. The fact that extraposed sentential subjects allow extraction
more easily might point to the fact that they arein fact base-created ascomplements,
as argued for by Webelhuth (1989). Recall, however, that this is a much smaller
effect than the subject/object asymmetry. The residual degradation compared to
objects could be related to the different matrix predicates. Alternatively, we could
follow B?uring & Hartmann (1997) and assume that extraction always orginiates
from the base position of sentential subjects, while extraposition is a post-syntactic
operation.
4.3.6 Conclusion
What we take away from this experiment is that we find a strong subject/object
asymmetry, no matter whether the non-finite sentential arguments are extraposed or
not. This again leads us to conclude that object, i.e., complements, are the preferred
extraction domains. Let us now turn to some other factors that have been argued
in the literature to influence the acceptability of these constructions.
4.4 Experiment 6 - Auxiliaries vs. Main verbs
4.4.1 Introduction
In our overview of the discussion on German subject islands in the theoretical liter-
ature in section 4.2, we saw that Grewendorf (1989) adds an intricate piece of data
he takes to be the crucial factor in deciding whether German allows wh-extraction
out of non-finite subjects or not. His observation was given in (3), repeated as (8).
135
(8) a. Wessen
whose
Beispiele
examples
hat
has
zu
to
analysieren
analyze
dich
you
mehr
more
frustriert
frustrated
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
?(*?) Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
b. *Wessen
whose
Beispiele
examples
frustrierte
frustrated
dich
you
zu
to
analysieren
analyze
mehr
more
-
-
Haiders
Haider?s
oder
or
Sternefelds?
Sternefeld?s
?(*?) Whose examples did it annoy you more to analyze - Haider?s or
Sternefeld?s??
In (8-a) the V2 position of the main clause is filled with the auxiliary hat (?has?)
and the extraction of the wh-element wessen Beispiele (?whose examples?) out of
the sentential subject is acceptable. (8-b) only differs in that the matrix V2 posi-
tion is filled with the main verb frustrierte (?frustrated?) and the sentence is judged
ungrammatical by Grewendorf. He argues that this is the crux of the prior contro-
versy concerning the status of subject islands in German. While Haider (1983) uses
auxiliaries and judges examples of this kind to be good, Sternefeld (1985) uses main
verbs and reports these cases to be out.
This is a beautiful example for seeing the binary view on grammaticality at
work. Clearly, if the two sentences in (8) are presented side by side, and a German
speaker is asked to assign each sentence the label good or bad, acceptable or unac-
ceptable, well-formed or not well-formed etc. most speakers will find that there is
a contrast between the two, and deem (8-b) a little worse than (8-a). The linguist
will happily conclude that (8-a) is good while (8-b) is bad.
Recall from our discussion in section 2.2 that this reasoning can lead to a
skewed view of the facts. Shoving every sentence in either the ?good? or the ?bad?
drawer misses important aspects of the data. We are also interested in (i) the extent
136
to which two sentences differ from each other and (ii) how each of them compares
to other sentences. So Grewendorf?s observation is highly interesting and clearly
in need of an explanation, but the conclusions he draws from it are misleading:
German allows extraction out of subjects as long as the matrix C position is filled
with an auxiliary. He then constructs a complicated story in terms of government
(which we will not go into here) to account for this alleged grammatical difference.
What he does not check, however, is whether this contrast is specifically re-
lated to subject islands or extends to other islands. Most importantly, he does not
investigate whether this is a general property of question formation in German even
in non-island contexts. These would be highly relevant facts to collect to get a bet-
ter idea of whether we are dealing with a somewhat unexpected constraint in the
grammar of German, or whether this effect results from a different source such as
parsing. This is in no way to discredit Grewendorf?s work, which we think is full
to the brim with insightful observations and keen theoretical proposals. What the
reader should take away from this digression, however, is that the dichotomy doc-
trine of grammaticality can blind researchers and prevent them from asking further
questions as well as gaining a more thorough picture of a given phenomenon.
4.4.2 Design and Methodology
In our previous experiment we found that there is an asymmetry in acceptabil-
ity between extraction out of sentential subjects as compared to extraction out of
sentential objects. While the former are often judged acceptable when a binary
categorical distinction is imposed on the speakers, using a wider and more flexible
scale reveals that speakers perceive them as degraded as compared to the object
conditions. Grewendorf claims to have detected the crucial factor that marks the
threshold between grammatical and ungrammatical, i.e. extraction out of subjects
is grammatical when the V2 position is filled with an auxiliary, and ungrammatical
137
when it is occupied by a main verb. In what follows, we will subject this claim to
further scrutiny.
We tested whether this is a peculiarity of extracting out of subjects, or whether
it also holds for extraction out of objects. The experiment had a 2x2 design with
the factors Sub/Obj and Main/Aux, which gives us the following stimuli:6
(9) a. Welche
which
Sonate
sonata
hat
has
denn
PRT
den
the.ACC
Pianisten
pianist
zu
to
spielen
play
gelangweilt?
bored
?Which sonata did it bore the man to play?? [sub, aux]
b. Welche
which
Sonate
sonata
langweilt
bored
denn
PRT
den
the.ACC
Pianisten
pianist
zu
to
spielen?
play
?Which sonata did it bore the man to play?? [sub, main]
c. Welche
which
Sonate
sonata
hat
has
denn
PRT
der
the
Pianist
pianist
zu
to
spielen
play
geplant?
planned
?Which sonata did the pianist plan to play?? [obj, aux]
d. Welche
which
Sonate
sonata
plant
plan
denn
PRT
der
the
Pianist
pianist
zu
to
spielen?
play
?Which sonata does the pianist plan to play?? [obj, main]
23 native speakers of German were tested. The by now familiar experimental pro-
tocol was followed.
4.4.3 Results
As in our previous experiments we find a strong main effect for the factor Sub/Obj,
i.e. extraction out of objects is preferred over extraction out of subjects. Grewen-
dorf?s generalization, according to which a main verb in the V2 position decreases
acceptability in the case of extraction out of subjects, is also confirmed. Further-
more, we find that there is a marginally significant main effect for the factor Main-
6(9-d) was changed into present tense since simple past tense ?plante? is rarely used in spoken
varieties of Austrian German, the dialect most of the participants spoke.
138
Figure 4.2: Effects of element in V2 position on extraction
Aux across the board. A pairwise comparison shows that the main vs. auxiliary
difference is only significant for subject (t(1,69)=1.81, p = .036) but not for objects
(t<1).
Interestingly, we do not find an interaction effect between the two factors,
i.e. using main verbs leads to a decrease both with subjects and objects and does
not affect one condition disproportionally. This is summarized in Figure 4.2, the
descriptive statistics in Table 4.5 and the ANOVA results in Table 4.6.
4.4.4 Discussion
Our results indicate that, ceteris paribus, the acceptability indeed drops if the aux-
iliary in the V2 position of the matrix clause is replaced with a full verb. While
we observe this trend across the board it only reaches signifiance in the case of
extraction out of subjects.
139
Table 4.5: Descriptive Statistics
Subject Object
auxiliary 3.49 5.64
main 2.88 5.38
Table 4.6: ANOVA
F p
Sub/Obj 125.2 ***
Main/Aux 7.6 *
Sub/Obj x Main/Aux .931 .338
We also find that the difference in acceptability between the main vs. auxiliary
condition with subjects (the dark grey column in Figure 4.2) is much smaller than
the overall difference between subjects and objects (the dark grey vs. the light grey
bars). It is quite remarkable that Grewendorf (Haider 1993 follows his judgments)
places the threshold between absolute grammaticality and absolute ungrammatical-
ity exactly between those two conditions, given that the effect is relatively small.
This seems like a rather arbitrary decision.
What can we make of the fact that there is only a significant difference between
auxiliaries and main verbs in the subject case? Note that we do see a small trend in
favor of auxiliaries even in the object case (5.64 vs. 5.38). A plausible interpretation
might be that this difference is amplified in the subject conditions. The presence
of another decreasing factor, i.e. the subject islandhood, makes the aux/main verb
difference more visible. We will see something similar with extraction out of subjects
in English in section 5.5.
What is the source of this aux vs. main verb difference? We can then resort
to some sort of parsing or processing story, arguing that having the main verb
in second position somehow makes it more difficult to retrieve the filler when we
reach the gap site because there is a cost for storing the verb in memory. The exact
140
opposite prediction, however, according to which knowing the verb and its argument
structure early could facilitate the retrival of the wh-element is conceivable as well.
We will have to leave this an open question. Suffice it to say, that if either processing
story is the right path to take it seems even more arbitrary to place an absolute
grammaticality threshold between extraction out of subjects with an auxiliary and
extraction out of subjects with a main verb.
4.4.5 Conclusion
To sum up the lessons learned from the experiment, we have reasons to believe
that an account in terms of parsing is most likely to explain these effects, given
that we see similar trends for both subjects and objects. But what exactly about
parsing causes this asymmetry? (9-a) and (9-b) (just as (9-c) and (9-c)) differ in
two respects: (i) the element that fills C, (ii) whether the predicate is separated or
not. Either of these factors could be responsible for the difference in acceptability.
The following experiment tries to tease these two differences apart and leads us to
conclude that factor (i) but not factor (ii) is responsible for the effect.
4.5 Experiment 7 - Separable verbs
4.5.1 Introduction
The purpose of this experiment is to follow-up the results of the study discussed
in the last section, which showed that speakers disprefer main verbs in the V2
position when a wh-element is extracted. We hypothesized that this is an effect
related to parsing. Now we would like to identify whether this is related to the fact
that having an auxiliary in the second position implies having a two-part predicate.
It is conceivable that the reoccurrence of the predicate close to the gap somehow
facilitates parsing. If this is the case, we would expect the same facilitation with
141
other two-part predicates, such as separable verbs.
This explanation is also contemplated by Haider (1993:159), who argues that
the parsing difficulty should disappear with separable particle verbs. He does not
give minimal pairs but judges extraction out of subjects involving particle verbs
with a (?), while assigning ?? to different examples with non-separable verbs.
(10) a. ??In
in
welche
which
Akten
files
verwehrte
denied
Einsicht
access
zu
to
nehmen
take
dir
you.DAT
der
the
Richter?
judge
?In which files did the judge deny you access??
(Haider 1993, 159: ex. (81b), my glosses, JJ)
b. (?)Worueber
about.what
fiel
fell
mit
with
dir
you.DAT
zu
to
plaudern
chat
keinem
noone.DAT
ein?
in
?What did chatting about with you occurred to nobody?
(Haider 1993, 159: ex. (83a), my glosses, JJ)
This is another domain where purely introspective gathering of judgments reaches
its limits. Haider is careful to qualify his statement saying that the parsing burden
?sollte? (?should?) disappear with particle verbs. Then he goes one to give a contrast
between (?) and ?? between two examples that are not minimal pairs. At no point
does he explicate what (?) or ?? mean in a purely binary view on grammaticality.
It is the purpose of this study to give some content to the question marks and to
test whether the parsing burden only should disappear or actually does.
4.5.2 Design and Methodology
Since we already established that the main/auxiliary asymmetry is found across the
board irrespective of the extraction site, we only concentrate on extraction out of
subject islands. We had a 2x2 design manipulating the factors Separable and
Extraction, which yields the following conditions:
142
(11) a. Die
the
Arbeit
paper
zu
to
schreiben
write
regt
annoys
die
the
Studentin
student
auf.
up
?Writing the paper annoys the student.? [+sep, -ext]
b. Welche
which
Arbeit
paper
regt
annoys
denn
PRT
zu
to
schreiben
write
die
the
Studentin
student
auf?
up
?Which paper does writing annoy the student?? [+sep, +ext]
c. Die
the
Arbeit
paper
zu
to
schreiben
write
?argert
annoys
die
the
Studentin.
student
?Writing the paper annoys the student.? [-sep, -ext]
d. Welche
which
Arbeit
paper
?argert
annoys
denn
PRT
zu
to
schreiben
write
die
the
Studentin?
student
?Which paper does writing annoy the student?? [-sep, +ext]
The verb aufregen (?annoy?) in (11-a) and (11-b) is a separable verb, a Partikelverb
in descriptive German grammar. Whenever the verb is raised to the V2 position, the
particle obligatorily stays low in V creating a Verbalklammer (?verbal bracketing?).
In (11-c) and (11-d) inseparable verbs are used. If verbal bracketing facilitates
parsing, and the particle somehow aids the speaker to reconstruct the verb in its
thematic position and thus also helps associate the sentential argument with the
verb, we would expect to see a boost in acceptability as compared to the control
condition where no such parsing aid is available. This hypothesis, however, is not
borne out by the facts.
4.5.3 Results
Unsurprisingly, we found the usual main effect for the factor Extraction. How-
ever, we found neither a main effect for the factor Separable nor an interaction
effect betwenn the two factors, as indicated by the almost perfectly parallel lines in
Figure 4.3. We find that non-separable verbs are preferred by speakers across the
143
gid19gid1
gid20gid1
gid21gid1
gid22gid1
gid23gid1
gid24gid1
gid25gid1
gid17gid5gid15gid12gid10gid2gid4gid13gid8gid7gid1 gid26gid5gid15gid12gid10gid2gid4gid13gid8gid7gid1
gid17gid11gid5gid9gid2gid10gid2gid3gid6gid5gid1
gid26gid11gid5gid9gid2gid10gid2gid3gid6gid5gid1
Figure 4.3: Results
board. Pairwise comparisions indicate that there is a marginally significant effect
for the +extraction conditions (t(1,62)=1.49, p=.069) and only a non-significant
trend in the -extraction conditions (t(1,62)=1.01, p=.15)). See also the descriptive
statistics in Table 4.7 and the ANOVA results in Table 4.8.
Table 4.7: Descriptive Statistics
Mean SD
-Sep, -Ext 6.15 1.62
-Sep, +Ext 2.67 1.61
+Sep, -Ext 5.84 1.78
+Sep, +Ext 2.23 1.58
Table 4.8: ANOVA
F p
Separable 2.5 .116
Extraction 385.7 ***
Sep x Ext .029 .865
4.5.4 Discussion
The lack of a main effect for the factor Separable and of an interaction effect be-
tween the two factors, suggests that verbal bracketing does not have any facilitating
effect for the parser, contrary to Haider?s hypothesis. We even see the opposite pat-
tern, i.e. non-separable verbs receive better ratings. The fact that this is signifcant
144
only for the +extraction conditions is reminiscent of what we found in the previous
experiment. An effect is amplified when it occurs at a lower range on the scale.
What seems to be responsible for the contrast between main verb and aux-
iliaries observed by Grewendorf and confirmed in the previous study is simply the
position of the element that bears the semantic content. Jeff Lidz (p.c.) suggests
that, in the main verb condition, storing the verb until the gap in the thematic
position is reached puts a burden on the working memory . The parser then needs
to reconstruct the trace of the head movement and associate it with the sentential
argument. The reconstruction step is not needed in the auxiliary conditions, since
the past participle that bears the semantic content of the predicate is adjacent to the
sentential argument. No reconstruction of the head movement is needed. Having a
verbal particle adjacent to the gap does not seem to felicitate the association with
the verb, since reconstruction of the main verb still needs to take place.
The parsing account could lead to the inverse prediction as well. It is con-
ceivable that having the thematic information early, i.e. when the main verb is in
C, helps the parser intergrate the arguments. Then we would expect a boost as
compared to seeing the main verb late. This is essentially an empirical question as
to which information is useful for the parser and which information is harmful. The
results reported in this section suggest that the parser does not like to see the verb
too early. However, frequency consideration - perhaps auxiliaries are more common
in interrogative C than main verbs - might also play a role. In short, much further
research in the processing of these constructions is needed to give a definite answer.
4.6 Conclusion
The experiments discussed in this chapter further fortified the view that objects
allow subextraction more easily than subjects. We have tested experimentally cer-
tain factors that were singled out in the theoretical literature as affecting the overall
145
acceptability. In particular, we looked at extraposition, main verbs vs. auxiliaries in
the V2 position and separable verbs. We thus aimed at providing a clean empirical
foundation for theoretical accounts to be built on. In some cases our experiments
have raised new questions about islands and wh-questions in general to be addressed
by a collaborative effort by syntacticians and psycholinguists. However, while not
all details of the acceptability patterns we found can be accounted for in terms of
complement/non-complement asymmetries, it seems fair to say that the CED goes
a long way towards capturing the basic patterns we found.
146
Chapter 5
Subextraction in English1
5.1 Background
In the first discussion of subject island, Ross (1967) covered a variety of cross-
linguistic data. His primary language of investigation, however, was English. Recall
from section 2.1.1.1 that he was careful to solely rule out extraction out of sentential
subjects to allow in examples like the following:
(1) Of which cars were the hoods damaged by the explosion?
Ross, 1967 (p.242, ex. 4.253)
Chomsky (1973) extended Ross?s original constraint to exclude extraction out of
subjects in general. After that, examples like (1) were buried in oblivion for a
couple of decades and are only enjoying a rennaisance in recent years. A small
industry is prospering which specializes in collecting prima facie counterexamples
to existing island constraints to disprove their existence. Subject islands have not
been spared the bad press either.
Levine & Sag (2003), who are concerned with ruling out extraction out of
subjects within the HPSG framework, mention the following examples, which they
find surprisingly good:
1The experiments discussed in sections 5.2, 5.3 and 5.4 are partially the result of a class project
at the University of Maryland in Fall 2009. Big thanks go to the students in this class: Jeff
Ackermann, Aleria Evans, Phil Glaser, Kassie Gynther, Grace Lavigne, Darren Samuels, Sarah
Slavin, Sherrod Wright.
147
(2) Levine & Sag 2003, fn. 6
a. (?)There arecertaintopics that jokes about arecompletely unacceptable.
b. (?)There are certain dignitaries that my jokes about are always consid-
ered over the top.
c. (?)There are certain dignitaries that my talking to would be considered
improper.
Sauerland & Elbourne (2002) claim that extraction out of passivized subjects in
English is acceptable under certain conditions. Concretely, they argue that the
raising of the subject to SpecTP can be delayed until PF and the subject can stay
in its complement position, if a quantifier is present which scopes over the subject.
Consequently, the subextraction is legitimate as it originates from a complement
position. This - according to Sauerland & Elbourne - is illustrated by the following
example:
(3) ?That?s the book that a different chapter of seems to have been assigned to
every student.
Beatrice Santorini is dedicating parts of her website2 to collecting real-life examples
of island violations. Here are some of the apparent subject island violations she
found attested:
(4) a. And a desert is one of those entities, like virginity and sans serif typeface,
of which the definition must begin with negatives.
(David Quammen. 1985. Natural acts. A sidelong view of science and
nature. New York: Laurel. 176.)
b. a letter of which every line was an insult
(Jane Austen. 1981. The complete novels. New York: Gramercy. 84.)
c. Their conversation turned upon those subjects of which the free discus-
sion had generally much to do in perfecting a sudden intimacy between
two young ladies
(Jane Austen. 1981. The complete novels. New York: Gramercy. 826.)
d. that voluminous publication, of which either the matter or manner would
not disgust a young person of taste
2http://www.ling.upenn.edu/?beatrice/examples/movement.html, last accessed April 21, 2010
148
(Jane Austen. 1981. The complete novels. New York: Gramercy. 828.)
Most recently, Chomsky (2008:20) has changed his opinion on this issue since Chom-
sky (1973:249) and resuscitated the debate by arguing that unaccusative, passive
and ECM subjects are not islands in English:
(5) a. it was the CAR (not the TRUCK) of which [the (driver, picture) was
found]
b. of which car did they believe the (driver, picture) to have caused a
scandal
It seems fair to say that the number of alleged counter examples to the subject
condition is worrisome for both theories of freezing and the CED. However, there
are a number of different issues with all of the above examples which have to be
disentangled before we can continue our discussion. First, what kind of position
are we actually subextracting from? Some of the examples above involve extraction
from passive and unaccusative subjects. We know from our experiments on German
that internal subjects allow extraction more easily than external subjects. However,
English, unlike German, is widely agreed to disallow in-situ subjects. If subjects by
necessity move to SpecTP in English, does it still matter whether the lower copy is
a complement or a specifier or do we unequivocally expect freezing effects?
Second, many of the examples above involve pied-piping of the preposition.
How can we be sure that genuine subextraction has taken place and that we are
not dealing with some kind of hanging topic construction? The issue becomes more
pressing when we slightly alter Santorini?s examples by stranding the preposition
(with all due apologies to Jane Austen). For most speakers, the acceptability of the
sentence decreases dramatically:
(6) a. ?*And a desert is one of those entities, like virginity and sans serif type-
face, which the definition of must begin with negatives.
149
b.?*a letter which every line of was an insult
c. ?*Their conversation turned upon those subjects which the free discussion
of had generally much to do in perfecting a sudden intimacy between
two young ladies
d.?*that voluminous publication, which either the matter or manner of would
not disgust a young person of taste
Third, how reliable are our guesstimates about the status of these examples? Do
they hold robustly across speakers? And most importantly, how do they contrast
compared to subextraction out of objects? This chapter is dedicated to shedding
light on these questions. In section 5.2 we will first investigate the role of Pied-Piping
with subextraction out of subject and object DPs. We will also take a closer look
in section 5.3 at whether ECM predicates somehow facilitate the extractability. We
will then turn our attention to the contrast between extracting out of unergative and
passivized subjects in section 5.4. Section 5.5 will revisit Sauerland & Elbourne?s
intriguing cases where subject condition effects allegedly disappear when the raising
of the subject can be delayed to PF for scopal reasons.
Following the practice of scientific papers of blatantly violating the rules of
dramaturgy, we will scotch every chance of tension build-up and give away the
bottom-line of our findings right away. Our results converge on the conclusion
that subjects are islands and uniformly opaque domains. Neither passives, ECM-
predicates nor obligatory scopal reconstruction can salvage them. The picture that
emerges with pied-piping seems to support the point of view that fronted PPs can be
construed as hanging topic constructions, which do not in fact involve subextraction
(as advocated by Broeckhuis 2005).
150
5.2 Experiment 8 - Pied-piping in English
5.2.1 Introduction
We saw in the the introduction above that many of the alleged cases of licit subject
subextraction involve Pied-piping of the preposition. We have noted that Ross in his
original discussion did not formulate a more general and simpler constraint banning
extraction out of subjects altogether but restricted it to sentential subjects. Let us
take a closer look at his key example, repeated here:
(7) Of which cars were the hoods damaged by the explosion?
Ross, 1967 (p.242, ex. 4.253)
This example is not innocent for a number of reasons. First, Ross chooses to extract
the the full PP and not NP stranding the preposition. Nothing in principle would
disallow this extraction since p-stranding is a productive phenomenon in English.
(8) ?*Which cars were the hoods of damaged by the explosion?
For many speakers stranding the preposition significantly decreases the acceptability
of the sentence. This is peculiar as most speakers of English generally prefer P-
stranding to pied-piping of the entire PP (barring prescriptive considerations).
(9) a. Which cars did the explosion damage the hoods of?
b. ?Of which cars did the explosion damage the hoods?
This suggests that ?? is not a case of genuine extraction but rather some sort of
hanging topic construction with an underlying structure such as (10).
(10) Of which cars was it the case that the hoods of those cars were damaged by
the explosion.
151
An argument for that construal is that, in an appropriate context, the DP the hoods
can be replaced by a pronoun. The fronted PP is still acceptable even though
pronouns do not normally take PP complements (a similar argument for Dutch is
made by Broekhuis 2005).
(11) There was a terrible explosion and the hoods of certain cars were damaged.
a. Of which cars were they damaged?
b. *Which cars were they of damaged?
c. *They of the SUVs were damaged.
d. The hoods of the SUVs were damaged.
In (11-a) we see that of which cars may still be fronted even if the subject of
the clause is pronominal. The impossibility of stranding the preposition in (11-b)
strongly suggests that the PP did not originate inside the subject but was base
generated in some left peripheral topic position. (11-c) illustrates that pronouns in
English do not take PP complements, while this is fully acceptable for regular DPs
as shown in (11-d).3
It seems like Ross chooses caution over simplicity in not proposing a general
subject constraint. A constraint that only applies to sentential subjects, however,
seems rather ad hoc and unlikely to be a foundational property of the grammar. It
seems that Ross might have been a little too wary of not ruling out examples incor-
3In addition, (7) is a passive sentence, and we generally assume that in passivization the overt
subject starts out as the thematic object. In the framework Ross was working in, passivization
is instantiated by an optional passive transformation rule. Nothing, in principle, would prevent
Ross from assuming that the WH-question rule precedes passive formation. In that case, at the
point of wh-extraction the overt subject would still be in object position, i.e. it would not be
immediately dominated by S and the transformation could apply. This step has to be allowed by
the system to generate extraction out of objects. Then the passive formation rule could apply,
reordering the thematic object into the syntactic subject position. The structural analysis of the
passive formation rule as given in Chomsky (1957) would have to be slightly modified, allowing
passive transformation to apply to WH - Aux - NP - V - NP. As Howard Lasnik (p.c.) points out,
however, that would run into empirical problems with sentences like Which boys were arrested by
Colonel Stumpi? If WH-question formation preceded passive formation, the plural agreement on
the verb would be unexpected. One could image optional ordering between the rules, but optional
rule ordering was not part of the Syntactic Structures system. We will return to extraction out of
passive subjects in section 5.4.
152
rectly. Genuine cases of extraction out of non-passivized unergative non-sentential
subjects are generally agreed on to be unacceptable:
(12) *Who did a book about cause a scandal?
The following experiment investigates the role Pied-piping plays both with subex-
traction out of subjects and objects. We will see that the results make a construal
of fronted PPs in terms of a base-generated hanging topic construction plausible.
5.2.2 Design and Methodology
We manipulate the factors Sub/Obj, Extraction and Pied-piping. This gives
us the following conditions:
(13) a. subject, no extraction
Phil wondered whether a documentary about healthcare had swayed the
voters last year.
b. subject, pied-piping
Phil wondered about which topic a documentary had swayed the voters last
year.
c. subject, no pied-piping
Phil wondered which politician a documentary about had swayed the voters
last year.
d. object, no extraction
Phil wondered whether Scott had filmed a documentary about healthcare
last year.
e. object, pied-piping
Phil wondered which topic Scott had filmed a documentary about last year.
f. object, no pied-piping
Phil wondered about which topic Scott had filmed a documentary last year.
As usual, we constructed 3 lexicalizations of every condition which were grouped
into 6 lists using a Latin square design. We consistently used d-linked wh-phrases
and extraction out of inanimate unergative subjects. We also added an adjunct at
the end of the clause (?last year? in (13)) to preempt a possible prescriptive bias
153
against (13-e), which would have ended with a preposition. The prescriptive rule
some speakers might remember as prohibiting ending sentences with a preposition
would bias against this condition. It would also make the interpretation of the data
more difficult as it would be unclear if the participants gave a rating according to
their intuitions or consciously applied a prescriptive rule.
We added 2 subexperiments as well as fillers of all level of acceptability, which
totalled in 82 items per participant. As usual, see the appendix for all stimuli and
fillers. 37 native speakers of English were tested online.
5.2.3 Results
The results we find give us an interesting picture. While there is a huge sub-
ject/object asymmetry when the preposition is stranded, there is only a small effect
with pied-piping. The descriptive results are given in Figure 5.1 and Table 5.1. The
black lines represent the object and the grey lines the subject conditions. The solid
lines are the pied-piping and the dashed lines the p-stranding conditions. We can
clearly see that the solid lines are nearly parallel indicating that there is a Sub/Obj
main effect but no (or only a very small)Sub/ObjxExtractioninteraction effect.
The dashed lines, however, are non-parallel and indicate a very strong interaction
effect.
Table 5.1: Descriptive results
no extraction +pied-piping -p-stranding
Subject 6.24 3.29 2.51
Object 6.54 3.86 5.08
Furthermore, two 2 way repeated measures ANOVA subanalyses were conducted,
which indicate that thereis ahighly significantSub/ObjxExtractioninteraction
if the preposition is stranded. Even though there also is an interaction effect with
154
gid17gid12gid16gid16gid1
gid18gid12gid16gid16gid1
gid19gid12gid16gid16gid1
gid20gid12gid16gid16gid1
gid21gid12gid16gid16gid1
gid22gid12gid16gid16gid1
gid23gid12gid16gid16gid1
gid13gid3gid10gid8gid1 gid24gid3gid10gid8gid1
gid7gid9gid2gid11gid1gid24gid6gid6gid1
gid7gid9gid2gid11gid1gid13gid6gid6gid1
gid5gid2gid4gid11gid1gid24gid6gid6gid1
gid14gid5gid2gid4gid11gid1gid13gid6gid6gid14gid1
Figure 5.1: The role of pied-piping in subject subextraction in English
pied-piping, the effect size is much smaller in the pied-piping conditions compared
to the p-stranding conditions. This is reflected in the drastic difference between
the ?2-scores (.733 vs. .184).4 This is also reflected in the pairwise comparisions.
Both are significant but the effect is larger for p-stranding compared to pied-piping
(t=2.36 vs. 11.65).
Table 5.2: 2x2 ANOVA subanalyses
Pied-piping P-stranding
F p ?2 F p ?2
Sub/Obj 140.566 *** .561 828.391 *** .883
Extraction 477.326 *** .813 698.022 *** .864
Sub/Obj*Extraction 24.849 *** .184 302.715 *** .733
4Remember that the partial ?2 measures the effect size, i.e. the proportion of the variance
that is accounted for by the independent variable. Following convention, .01 is considered a small
effect, .09 a medium effect, and .25 a large effect (cf. Sprouse 2007, Weskott & Fanselow 2008).
155
5.2.3.1 Distribution of the data
A more detailed look at the individual results suggests that they are representative
of one grammar. While we observe the usual amount of variation in the data,
25 participants follow the same general pattern as the mean and 12 participants
show a different pattern (i.e. about 68%). However, across the 12 participants who
did not follow the pattern, we could not detect a standard pattern. For example,
some participants rated obj, -pp as less acceptable than obj, +pp, instead of more
acceptable like most other participants. This might reflect an actual difference in
their grammars, but the judgment could also be confounded by the prescriptivist
rules on p-stranding.
The histograms in Figure 5.2 summarize the individual distribution of the
data. The charts on the right show the p-stranding conditions (subjects top, objects
bottom) and very clearly indicate a complementarity in the rating. The judgments
for the subject condition cluster in the bottom third of the scale while the object
condition curve has its highest density in the top third of the scale. The histograms
on the left representing the individual distribution of the Pied-piping condition, on
the other hand, are much more alike. The top chart show the subject and the
bottom chart the object conditions. Both ratings cluster around the middle of the
scale.
5.2.4 Discussion
A simplistic look at the p-values in Table 5.2 might lead us to conclude that there
is a Sub/Obj*Extraction for both pied-piping and p-stranding and that?s that.
However, we observe a striking asymmetry in that pied-piping decreases the accept-
ability in the object but increases it in the subject conditions, bringing the two
values much closer together. In other words, the subject/object asymmetry almost
goes away with pied-piping, a trend which is also reflected in the relatively small ?2
156
Figure 5.2: Pied-piping histograms
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid13gid14gid5gid3gid3
gid3
gid4gid3
gid5gid3
gid6gid3
gid7gid3
gid1gid12
gid7gid11
gid14gid7
gid9gid6
gid15
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid14gid15gid5gid9gid10gid3gid3
gid3
gid4gid3
gid5gid3
gid6gid3
gid7gid3
gid1gid13
gid7gid12
gid15gid7
gid9gid6
gid16
gid16gid22gid18gid2gid1gid12gid19gid23gid2gid1gid11gid1gid4gid2gid7gid8gid10gid9gid10
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid11gid5gid9gid3gid3
gid3
gid8
gid4gid3
gid4gid8
gid5gid3
gid5gid8
gid1gid14
gid7gid13
gid15gid7
gid10gid6
gid16
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid11gid5gid9gid10gid11gid3gid3
gid3
gid4gid3
gid5gid3
gid6gid3
gid7gid3
gid1gid14
gid7gid13
gid15gid7
gid10gid6
gid16
gid17gid23gid19gid2gid1gid13gid20gid24gid2gid1gid12gid1gid4gid2gid11gid8gid7gid7gid4
157
values.
We take this to be evidence for the position advocated for by Broeckhuis
(2005) for Dutch and also defended for English in 5.2.1 above. According to this
view, labelling the pied-piping conditions +extraction is a misnomer, as no genuine
subextraction has taken place. Instead, the PP was base-generated as some sort
of hanging topic or aboutness construction in the C-domain and no filler-gap de-
pendency is established. The cost of associating this PP with the NP is roughly
equivalent for subjects and objects. A prediction of this explanation would be that
making the hanging topic pragmatically more plausible by adding an appropriate
context should improve the acceptability. Informally, this predictions seems to be
borne out:
(14) A documentary about the economy really helped the cause of the Demo-
cratic party.
About which topic did a documentary help the Republians?
An analysis in terms of hanging topics is not available with p-stranding. The prepo-
sition is an overt marker of the gap position. Here we observe our familiar sub-
ject/object asymmetry. In (13-c) the extraction originated from a unergative sub-
ject position. The strong degradation comes as no surprise as both the CED and
freezing are violated. Note that this experiment, and all our studies in English in
general, do not allow us to tease apart non-complement effects from freezing effects.
As subjects in English obligatorily move to SpecTP, the two are conflated. We know
that this degradation is a results of the structural position of the extraction domain
rather than the internal structure of the NP. The same NPs allow subextraction if
they are situated in the complement position of V, as in (13-e).
158
5.2.5 Conclusion
We conclude that genuine extraction only takes place in the p-stranding conditions,
while the +pied-ping examples involve some sort of topicalized aboutness PP. We
arrive at this conclusion through the observation that subject/object asymmetries
largely disappear when pied-piping is involved. Unsurprisingly, our results confirm
the classic subject/object asymmetry with genuine subextraction out of NPs. As a
results, the alleged counterexamples to the subject condition in English reported in
the literature are not cases of genuine extraction and as such do not pose a problem
for theories of freezing or the CED.5
5.3 Experiment 9 - ECM
5.3.1 Introduction
We saw in section 5.1 that the opinions on the permeability of the complements of
ECM verbs have been somewhat changeable. The ? in (15) should be read as a
diacritic of confusion rather than a qualitative judgment about its acceptability.
(15) ?Which politician does John believe a book about to have caused a scandal.
Why should ECM predicates matter? It is well-known that the thematic subject of
the embedded clause is exceptionally assigned accusative case by the matrix verb,
hence the term Exceptional Case Marking (ECM). In a sense the NP ?a book about
which politician? is Janus-faced between being a subject and an object. It would
hence be conceivable that it acts as a complement in terms of its permeability, an
intuition pursued by Chomsky (2008), albeit in much more technical terms.
5It must be noted that Levine and Sag?s counterexamples. (2-c) was rated around 4.3 as a
filler item. Here a reanalysis in terms of hanging topics of some sort seems implausible, given
that the preposition is left in-situ. It might be relevant that their examples involve passive and
unaccusative predicates, even though we will see in section 5.4 that the predicate type does not
matter for NP-subextraction. We have to leave these cases open for future research.
159
However, the closer we look at the phenomenon the less clear it becomes why
ECM-NPs should allow subextraction. There is no doubt that the NP starts out
within the thematic realms of the embedded predicate. Case assignment and word
order then give us strong evidence that it raises to somewhere in the no man?s land
between the matrix V and the embedded T (presumably filled with ?to?). This
is not the place to determine what the exact position is, and thankfully it seems
immaterial for our purposes (see Lasnik & Saito 1991 for discussion). If freezing is
a real phenomenon - and there is a plethora of evidence that it is - the grammar
should not care about what the position of the higher copy is. All that matters is
that it is a higher copy, i.e. that movement has taken place. There is no reason why
the ECM-NP should differ in any way from a regular subject that raises to SpecTP
for EPP reasons. In both cases we expect this NP to be an opaque domain due to
freezing.6
The default and expected situation would be that ECM-NPs disallowed all
kinds of subextraction. This would be good news, as our theories of freezing would
trivially extend to them. If, however, ECM turns out to create more transparent
extraction domains this would require significantly more complicated theories of
freezing. It is thus of some importance to find out what the facts really are.
5.3.2 Design and Methodology
This experiment has a simple 2x2 design, manipulating the factor ECM and Ex-
traction. This gives us the following conditions:
6Norbert Hornstein (p.c.) points out that a distinction could be made in terms of whether the
movement occured purely for EPP reasons or whether some case feature is checked. This could
conceivably have consequences for the permeability of a domain. Such a distinction could certainly
be formalized but it is not immediately obvious why it should exist.
160
(16) a. -ECM, -EXT
A book about Obama caused a scandal.
b. -ECM, +EXT
Which politician did a book about cause a scandal?
c. +ECM, -EXT
John believed a book about Obama to have caused a scandal.
d. +ECM, +EXT
Which politician did John believe a book about to have caused a scan-
dal?
As usual, we constructed 3 lexicalizations of every condition which were grouped
into 4 lists using a Latin square design. We consistently used d-linked wh-phrases
and extraction out of inanimate unergative subjects. We added 2 subexperiments as
well as fillers of all level of acceptability, which totalled in 82 items per participant.
37 native speakers of English were tested online.
5.3.3 Results
Unsurprisingly, our results clearly indicate that both factors have a significant effect
on acceptability. There is a cost associated to adding in an ECM predicate which is
independent of the extraction. The descriptive results are summarized in Table 5.3
and Figure 5.3.
Table 5.3: Means
-extraction +extraction
no ECM 6.77 2.61
ECM 4.48 2.24
Table 5.4: ANOVA
F p
ECM 424.851 ***
Extraction 1764.356 ***
ECM x Ext 148.72 ***
Furthermore, the ANOVA in 5.4 indicates that there is also a significant ECM x
Ext interaction effect. The negative effect of extraction on the rating is dispro-
portionately lower for the ECM condition. This sort of evidence is often used to
161
gid12gid1
gid13gid1
gid14gid1
gid15gid1
gid16gid1
gid17gid1
gid18gid1
gid10gid5gid9gid8gid1 gid19gid5gid9gid8gid1
gid6gid7gid1gid3gid2gid4gid1
gid3gid2gid4gid1
Figure 5.3: Extraction out of ECM-marked subjects
make the case that the lower drop of the acceptability indicates that this domain
is more permeable compared to the control condition. We have to be very careful
in this case though. Both extraction conditions and are rated very low on the scale
(2.61 vs. 2.24). A pairwise comparision shows that there is a marginally significant
difference: t(1,111)=1.87, p=.031).
5.3.4 Discussion
Our results do not give us much reasons to believe that ECM marked NPs are trans-
parent domains for subextraction. Even though there is a marginal improvement
for extraction out of ECM predicates, it is relatively close to the floor and we should
be careful in how to interpret it. The relatively large statistical interaction effect
we get is misleading as it is primarily a result of decreased acceptability of the base-
line condition caused by the complexity of a sentence such as (16-c). It is unclear
whether we want to invoke a grammatical account for this small effect.
This concurs with the majority of intuitions reported in the theoretical litera-
ture (Chomsky 1973, Diesing 1990). Chomsky?s (2008) examples, which are deemed
162
acceptable, have the confounding factor that they involve pied-piping. We have
shown in section 5.2, however, that there are reasons to believe that pied-piping
does not necessarily involve genuine subextraction. We will conclude for now that
ECM only marginally affects subextraction out of subjects.
The stimuli we used also differ in other respects. The ECM conditions are
longer and the the filler-gap distance in the ECM case is bigger than in the non-
ECM conditions. A straightforward follow-up study that would not have these issues
could compare ECM marked NPs with subjects of finite clauses.
(17) a. Which politician did John believe a book about to have caused a scan-
dal.
b. Which politician did John believe that a book about has caused a
scandal.
5.4 Experiment 10 - Passives
5.4.1 Introduction
We know from a German that internal and external subjects show different behavior
with regards to subextraction. Recall that we found ina number of studies in chapter
3 that subjects of unaccusatives and passives are more permeable than subjects of
unergatives. Should we now expect that English shows a similar asymmetry, as
argued for by Chomsky (2008)?
We know that German and English differ in that only the former allows for
subjects to stay in-situ. The question is whether the structural position of the lower
copy has an affect on the acceptability of subextraction. In other words, in a tree
like (18), does the grammar care about whether the NP in SpecTP originated from
the Spec of vP or the complement position of V?
163
(18) TP
NPi vP
< NPi > VP
V < NPi >
If we believe in some version of the cycle we expect the raising of the NP to SpecTP
to precede the subextraction out of it. The C head above TP then probes down its
c-command domain looking to satisfy its wh-feature. Under virtually any definition
of minimality the NP in SpecTP is the closest and as such the one that is attracted.
If we follow these assumption the structural position of the lower copy should be
immaterial.
However, the position of the lower copy becomes relevant under an MSO ap-
proach. Recall from section 2.1.2.1 that CED effects are a result of the specifier (or
adjunct) being forced to spell out early for linearization purposes. Consequently,
the specifier is turned from a hierarchical set to a string and its subcomponents are
no longer accessible for further syntactic movement.
For elements that are first-merged as specifiers this reasoning derives island
effects without further assumptions. Things get a little trickier with first-merged
complements that are raised to a Spec position, such as passive subjects in English.
Consider the following example, taken from Nunes & Uriagereka (henceforth NU,
2000, 26, ex. 17a ):
(19) *whoi was [ a picture of ti ]k taken tk by Bill
What would the prediction of MSO be? The DP ?a picture of who? is assembled and
merged with V. T is merged and requires a DP to move to its Spec to have its EPP
feature checked. ?A picture of who? is copied, spelled-out and remerged in SpecTP.
164
The following object is formed (the angled brackets indicate a spelled-out copy).
(20) [TP [DP <a, picture, of, who> ] [T were [V P taken [a pictures of who ] by
Bill ]]]
[NU?s ex. (21), slightly modified]
In the next step C is merged and probes down the spine to have its wh-feature
checked. The substructure of the higher copy in SpecTP, <a, picture, of, who>,
is not accessible. But what about the lower copy in the complement position of
V? Nothing up to this point prevents C from probing the lower ?who?, copying and
remerging in SpecCP, deriving (19).
NU are very much aware of this state of affairs and add the assumption that
uniform chains can only be formed between copies of the same type, i.e. a spelled-
out and a non spelled-out copy cannot form a legitimate chain. This forces the lower
copy to be spelled-out as well, when the higher copy is merged in SpecTP.
NU quite explicitly introduce this extra assumption to rule out examples like
(19). We will not question the naturalness of this account (see Sheehan 2009 for
discussion) but we will ask the more general question of whether it is empirically
desirable to fully rule out such examples. If the structural position of the lower
copy really plays a role, as contended by Chomsky (2008), the MSO system might
make a correct prediction without NU?s extra assumption, i.e. freezing effect are
ameliorated if the lower copy is a complement.
To this end, we compare the acceptability of extraction out of unergative and
passive subjects experimentally.
5.4.2 Design and Methodology
Just like the previous ECM-experiment this study has a simple 2x2 design, ma-
nipulating the factors Unergative/Passive and Extraction. This gives us the
165
following conditions:
(21) a. Unergative, +ext
John wondered which man a book about caused a scandal last year.
b. Unergative, -ext
John wondered whether a book about Obama caused a scandal last
year.
c. Passive, +ext
John wondered which man a book about was released last year.
d. Passive, -ext
John wondered whether a book about Obama was released last year.
As usual, we constructed 3 lexicalizations of every condition which were grouped
into 4 lists using a Latin square design. We consistently used d-linked wh-phrases
and extraction out of inanimate NPs. We added 2 subexperiments as well as fillers of
all level of acceptability, which totalled 82 items per participant. 37 native speakers
of English were tested online.
5.4.3 Results
The sample means in Table 5.5 andthe graph in Figure5.4 suggest that the predicate
type does not seem to have any large effect on acceptability. As a matter of fact,
the lines are close to being identical.
Table 5.5: Means
-ext +ext
Unerg 6.27 2.55
Pass 6.21 2.68
Table 5.6: ANOVA
F p ?2
Arg 1.491 .225 .013
Extraction 929.444 *** .894
Arg x Ext 15.056 *** .12
The ANOVA results in Table 5.6 confirm that there is a highly significant Extrac-
tion but no Arg main effect. There is, however, a small interaction effect between
the two factors. This interaction has to be taken with a grain of salt though. Even
166
gid16gid1
gid17gid1
gid18gid1
gid19gid1
gid20gid1
gid21gid1
gid22gid1
gid14gid4gid12gid8gid1 gid23gid4gid12gid8gid1
gid2gid3gid9gid11gid4gid1
gid6gid2gid7gid7gid5gid11gid4gid1
Figure 5.4: Extraction out of passivized subjects
though the p-value indicates significance, the effect size expressed by the partial
?2 statistic is very small. In addition, the pairwise comparision for the extraction
conditions (21-a) and (21-c) does not reach significance (t<1). A simple one-way
ANOVA comparing the means of the two extraction conditions also did not come
close to significance F(1,110) = .3694, p = .544). In short, even though we find a
small interaction effect we should very careful about the conclusions we draw.
5.4.4 Discussion
It seems fair to conclude that our experiment has confirmed the null-hypothesis that
the predicate type does not affect acceptability of subextractions out of subjects in
English. We argued above that this is expected if only the higher copy in SpecTP
is considered. The structural position of the lower copy is immaterial.
These results stand in contrast to findings cited by Kratchenko, Xiang & Polin-
sky (2009, henceforth KXP), who report an unergative/unaccusative asymmetry for
extraction out of subjects in English both in o?ine acceptability and self-paced
reading tasks.
167
KXP?s and the study presented here differ in a few points. The most important
difference is that, while both experiment look at internal vs. external arguments,
we were using passives and KXP unaccusative predicates. It would be interesting
to look at a four way comparison between unergatives, unaccusatives, passives and
objects, as in the German study discussed in section 3.4.7 We leave this for future
research.
Both our results and KXP?s show acceptability rating at the very low end of
the scale for both internal and external arguments, in striking contrast to the results
we got for German where internal subject were on par with objects. Even though the
contrast KXP find between unaccusative and unergatives is intriguing and in need
of an explanation, although not necessarily a grammatical one, it is overshadowed
by the contrast with objects. I take both our study and KXP as indicative of the
existence of freezing effects in English, with the structural position of the lower copy
being of only secondary importance.
5.5 Experiment 11 - Subject islands and reconstruction
5.5.1 Introduction
Sauerland & Elbourne (2002, henceforth SE) add an intriguing piece of data to
the discussion on subject islands. In essence, their account of scope reconstruction
predicts that the raising of internal arguments to SpecTP in passive clauses can
be delayed until PF if this postponement has a scopal effect. More technically, SE
restate Fox?s (1995) condition on total reconstruction in the following way:
7There are some other small differences that are unlikely to have a major effect on the results.
KXP used non-d-linked while we were using d-linked wh-elements. This difference is likely to
influence the Extraction main effect but unlikely to alter the relative contrast between the
condition (cf. Sprouse 2007). We used a 7-point and KXP a 5-point scale. With a relatively high
n (both studies had 37 people), this is also extremely unlikely to have any interesting effect on the
results (see our discussion in section 2.2.4.1).
168
(22) Overt movement of XP can be delayed until PF only if there is a scope-
taking element Y such that XP takes scope above Y if movement takes
place in the stem but below Y if movement is delayed until PF, and if these
two scopal construals are semantically distinct.
(SE, p. 303, ex. 54)
SE report personal communication with J. Nissenbaum, who claims that extraction
out of passive or unaccusative subjects is possible if the movement from the internal
argument position to SpecTP can be delayed according to (22). SE contend that
this is in fact borne out (SE, p. 304, ex. 56):
(23) a. *That?s the book Opj that [a chapter of tj ]i seems t?i to have been
assigned to John ti .
b. ?That?s the book Opj that [a chapter of tj ]i seems t?i to have been
assigned to every student ti .
In (23-a) the raising of ?a chapter of OP? has no reason to be delayed as ?John?
is not a scopal element. Hence, the movement happens cyclically in narrow syn-
tax and subextraction is ruled out as a result of freezing. In (23-b), on the other
hand, ?a chapter of OP? reconstructs within the scopal domain of ?every student?,
i.e., interpreting it in its in-situ position leads to semantically distinct construal
according to (22). Consequently, the movement can be delayed to PF and follows
the subextraction. The OP is subextracted out of a complement and the sentence
is grammatical.
This is arguably one of the most intriguing contrasts observed in the domain
of subject islands and a beautiful piece of evidence in favor of SE?s PF-movement
analysis. In fact it is such a good argument that we might start getting worried.
After all, if it seems too good to be true it probably is, as the proverb goes. The
logic of the SE?s theoretical argument seems sound but the question remains if the
contrast in (23) in fact holds as stated.
169
If linguists who are also native speakers of English are confronted with (23)
interesting things start happening. If we omit SE?s judgments and ask whether
they find a contrast between (a) and (b), none of my informants gets any difference.
Both are equally degraded, ?* is the consensual diacritic used. If we, however, first
explain SE?s theory and how it makes this prediction and then ask whether they
agree with the judgment given in (23), for many the contrast all of a sudden starts
to be real. It seems like this is the kind of contrast linguists would really like to
hold. We certainly do not intend to rain on SE?s parade but it seems quite crucial
to establish that this contrast holds up in a controlled acceptability judgment study
before it can used as evidence for a particular theoretical position.
5.5.2 Design and Methodology
The study has a 2x2 design, manipulating the factors Quantifier and Extrac-
tion. We also added a control condition where the NP that is subextracted out of
is located in an object position. This condition was added to make sure that the
NPs allow subextraction at all. This gives us the following five conditions:8
8Special thanks goes to Brad Larson for proof-reading the stimuli
170
(24) a. -Quantifier, -Extraction
A different chapter of the textbook seems to have been assigned to
John.
b. +Quantifier, -Extraction
A different chapter of the textbook seems to have been assigned to
every student.
c. -Quantifier, +Extraction
That?s the book that a different chapter of seems to have been assigned
to John.
d. +Quantifier, +Extraction
That?s the book that a different chapter of seems to have been assigned
to every student.
e. Control
That?s the book that John seems to have assigned a different chapter
of to every student.
As usual, we constructed 3 lexicalizations of every condition which were grouped
into 4 lists using a Latin square design. We followed SE in using presentational
relative clauses for all extraction condition. The adjective ?different? was added to
every condition to strongly favor a narrow scope reading in (24-b), (24-d) and (24-e),
i.e. the relevant reading that involves reconstruction to the in-situ position.9
One unrelated subexperiment as well as fillers of all level of acceptability were
added, which totalled 64 items per participant. 31 undergraduates of the University
of Maryland (all self-reported native speakers of English) took the experiment on-
line for class credit (Introduction to Linguistics, no participant had any other prior
training in linguistics).10
9Needless to say, the wide scope reading is still available but strongly dispreferred. We will
return to this issue below.
10Valentine Hacquard (p.c.) suggests a follow-up experiment contrasting two quantifiers with
the same quantificational force:
I i That?s the book that a different chapter of seems to have been assigned to some
student.
ii That?s the book that every chapter of seems to have been assigned to every student.
In (24-i) and (24-ii), there would be no delay until PF to raise since there would not be any scope
difference, i.e. the inverse scope would be logically equivalent. It would be very interesting to see
171
gid24gid20gid23gid23gid1
gid25gid20gid23gid23gid1
gid26gid20gid23gid23gid1
gid27gid20gid23gid23gid1
gid28gid20gid23gid23gid1
gid29gid20gid23gid23gid1
gid30gid20gid23gid23gid1
gid21gid8gid18gid14gid1 gid31gid8gid18gid14gid1
gid2gid12gid1gid4gid17gid5gid11gid15gid9gid8gid13gid1
gid4gid17gid5gid11gid15gid9gid8gid13gid1
gid3gid6gid10gid8gid7gid14gid1
Figure 5.5: Results
5.5.3 Results
The descriptive results are given in Figure 5.5 and Table 5.7. The solid grey line
represents the conditions without a quantifier, (24-a) and (24-c), the dashed grey
line the conditions with a quantifier, (24-b) and (24-d). The solid black line is the
object control condition.11
The 2x2 ANOVA subanalyses results are summarized in Table 5.8. While we
do find a huge effect for the factor Extraction and a large effect for the factor
Quantifier, we find a significant yet small interaction between the two factors
(the effect size is indicated by the ?2 value). A pairwise comparison between the
two subject subextraction conditions (24-c) and (24-d) is marginally significant (t(1,
92)=1.53, p=.063).
if and how this would affect the extraction out of the subject.
11Note that, technically, this should not be a line since we did not measure the object baseline
condition as this was not what we were primarily interested in. The object subextraction condition
was only added as a control to make sure that the NPs we were using generally allow subextraction.
We will not draw any conclusions as to the difference between subjects and objects in this study. It
is only to facilitate the optical comparison with the two subject lines that we assumed an equivalent
baseline value for the object as for the +quantifier, -extraction condition. Strictly speaking, this
is not fully kosher but we will hope for the reader?s clemency.
172
Table 5.7: Means
-ext +ext
no quantifier 6.18 2.88
quantifier 6.37 3.25
object N/A 4.44
Table 5.8: ANOVA
F p ?2
Quan 59.542 .225 .393
Ext 695.287 *** .883
Quan x Ext 8.105 *** .081
5.5.3.1 Distribution of the data
19 out of 31 participants (about 61%) confirm SE?s contrast between (24-c) and
(24-d). However, only 7 speakers (about 22%) rate this contrast with 1 point or
more on the scale. The complete distribution is given in the histograms in Figure
5.6.
Figure 5.6: Histograms
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid7gid8gid10gid13gid3gid7gid5gid14gid12
gid3
gid8
gid4gid3
gid4gid8
gid5gid3
gid5gid8
gid1gid11
gid5gid10
gid13gid5
gid7gid4
gid15
gid4gid2gid3gid3 gid5gid2gid3gid3 gid6gid2gid3gid3 gid7gid2gid3gid3 gid8gid2gid3gid3 gid9gid2gid3gid3 gid10gid2gid3gid3
gid9gid12gid3gid7gid5gid13gid11
gid3
gid8
gid4gid3
gid4gid8
gid5gid3
gid5gid8
gid1gid10
gid5gid9
gid12gid5
gid7gid4
gid14
gid17gid23gid19gid2gid1gid13gid20gid24gid2gid1gid12gid1gid4gid2gid9gid4gid11gid8gid8
We see the different means reflected in the slight tendency to the left between
the -quantifier condition compared to the +quantifier condition. However, there
are no reasons to believe that we are conflating different grammars. The data is
173
relatively uniform and shows the noise we used to seeing with acceptability judgment
tasks.
5.5.4 Discussion
Do our results answer the question whether SE?s contrast holds empirically? Yes and
no. At a first glance, the graph in Figure 5.5 presents us with two almost perfectly
parallel lines for the subject subextraction conditions. Speakers generally prefer the
presence of a quantifier in the object position. This is not surprising, given that the
adjective ?different? biases towards an inverse scope reading. The preferred meaning
in (24-b) is one where the universal quantifier scopes over the existential, i.e. the
distributed reading where chapter 1 is assigned to Mary, chapter 2 is assigned to
Grace, chapter 3 to Phil, etc. Pragmatically, this reading seems plausible, as we
can easily think of a seminar scenario where every student gives a presentation on a
different chapter of the textbook and we do not expect to listen to 20 presentations
on the same chapter.
Once the quantifier is replaced with a proper name, as in (24-a), we only have
one reading, namely the one where John was assigned a chapter of the textbook
that is different in some contextually salient way. No information is given what
this chapter is different from and the speakers have to construct a plausible context
themselves. This is certainly doable, e.g. John was originally assigned to present
chapter 3 but the teacher decided to skip this chapter and assigned John chap-
ter 4 instead. However, this extra pragmatic assumption seems to be enough to
bias against this condition. This is reflected in the different rating in the baseline
condition: 6.37 with the quantifier, 6.18 without it.
We find the same difference for the extraction conditions, (24-c) vs. (24-d).
However, the lines are not completely parallel. It seems to be the case that, even
when we control for the Quantifier main effect, extraction disproportionally de-
174
grades the -quantifier condition more than the +quantifier condition. This is exactly
the contrast reported by SE. The effect, however, is very small (?2=.081, following
convention effects < .09 are considered small effects). In addition, if we compare
the difference between the means of the two subject subextraction conditions, on
the one hand, to the difference between the +quantifier condition and the object
subextraction control condition, on the other hand, we see a striking discrepancy in
size: .37 vs. 1.19.
A reasonable question to ask is: is the difference caused by the presence of
the quantifier aggravated at a lower end of the scale? Speakers disprefer having a
proper name to having a quantifier for the pragmatic reasons explicated above. Once
extraction pushes down both conditions to the lower end of the scale the perceived
contrast aggrandizes. Does that super-additive effect mean that the presence of the
quantifier genuinely helps the subject subextraction in the sense of SE?
Consider this analogy: someone is asked to compare the taste between Coke
and Pepsi. First, both drinks are served ice cold and the participant has a slight
preference for Coke over Pepsi. Now, both sodas are served warm. The overall rating
drops dramatically, as expected, because sodas are supposed to be consumed chilled.
In addition, the perceived inferiority of Pepsi is larger than in the cold condition.
Does that mean that the factors Temperature and Soda Type interact? Is there
something about Coke that tolerates lukewarm temperatures better than Pepsi? Or
is it just the case that a perceived contrast becomes more noticeable at the lower
end of the scale? Intuitively, I am leaning towards the latter interpretation, both in
the Pepsi challenge and in the interpretation of our results above.
The difference between a true interaction effect and the magnification of an
effect at the lower end of the scale is subtle, and we will not be able to tease apart
the two here. It seems fair to conclude, however, that our data is nowhere near
being knock-down evidence for a grammatical contrast in the sense of SE. In their
175
discussion, they fail to give the full paradigm including object subextraction. Their
theory would predict that if subject raising can be delayed until PF, subjects should
essentially behave like objects for the purposes of subextraction. This is clearly not
the case. Whether there is a difference between the two subject conditions, related
to the presence of a quantifier, can neither be fully confirmed nor disproved by this
study.
5.5.5 Conclusion
This experiment was aimed at testing the correctness of SE?s claims that the pres-
ence of a scopal element affects the acceptability of subject subextraction. Our
experiment does not give a definite answer but we saw reasons to doubt that SE?s
contrasts holds as presented in their paper. Given the small effect size, the base-
line difference and the contrast with the object control condition, the conservative
interpretation of the results disfavors interpreting this contrast as indicative of an
architectural difference in the sense envisaged by SE.
176
Chapter 6
A glance at other languages: Subextraction in
Japanese and Serbian
6.1 Japanese
6.1.1 Introduction
Japanese is often reported to allow extraction out of subjects (cf. Ross 1967, Ishii
1997, Takahashi 1994, Kikuchi 1987 among others). Stepanov, citing Kikuchi, gives
the following contrast for comparative deletion:
(1) a. Op [Mary-ga
Mary-nom
t yonda
read
no]-ga
that-nom
akirakana
is-obvious
yorimo
than
John-wa
John-top
takusan-no
many-gen
hon-o
book-acc
yonda.
read
?(*)John read more books than [that Mary read t ] is obvious.?
b. *Op Bill-ga
Bill-nom
[Mary-ga
Mary-nom
t yonda
read
kara]
because
odoroita
was-surprised
yorimo]
than
John-wa
John-top
takusan-no
many-gen
hon-o
book-acc
yonda.
read
?(*)John read more books than Bill was surprised because Mary read t.?
In (1-a) the operator originates inside a subject and the construction is reported to
be well formed. In (1-b), on the other hand, the operator-gap dependency is formed
across an adjunct, which is deemed unacceptable. Stepanov takes the existance of
adjunct island effects as indicative of the formation of a genuine filler-gap depen-
dency. He argues against an analysis in terms of a silent pro, as otherwise there
177
would be no reason for (1-b) to be ill-formed. He then concludes that Japanese
allows extraction out of subjects.
There are at least two problems with Stepanov?s examples. First, the exam-
ples in (1) and virtually all other cases cited in the literature involve highly subtle
judgments, and some speakers disagree with what is reported. While all speakers
happily concur with the unacceptability of the adjunct case, speakers have very
mixed opinions about the subject case. Some even have problems understanding
what (1-a) is supposed to mean. In short, while there is undoubtedly a contrast be-
tween extraction out of subjects and adjuncts, classifying this asymmetry in terms
of * vs. ok seems too premature a conclusion.
Second, Stepanov looks at the wrong contrast to begin with. It is no surprise
that extraction out of adjuncts is degraded compared to extraction out of subjects.
The strength of islands effects is known to vary between different types of islands.
This does not imply that weaker island effects do not exist. The contrast between
subject and adjuncts holds for English as well, as the following contrast suggests.
The fact that (2-a) is slightly less degraded than (2-b) certainly does not imply
that the former is grammatical and the latter ungrammatical. What is relevant in
English is that subject subextraction is degraded compared to object subextraction.
(2) a. ?*Which politician did a book about cause a problem because it insulted
John?
b. *Which politician did John cause a problem because he wrote a book
about?
We conclude that Stepanov?s data is inconclusive. Another case is considered by
Lasnik&Saito (1992) (henceforth, LS), who to my knowledge are the only ones who
directly contrast extraction out of subjects with extraction out of objects.
178
(3) a. ??dono
which
hon-o
book-acc
Mary-ga
Mary-nom
[John-ga
John-nom
t katta
bought
koto]-o
fact-acc
mondai-ni
problem-to
siteru
making
no
?Which book is it that Mary is calling the fact that John bought into
question?
b. ??dono
which
hon-o
book-acc
Mary-ga
Mary-nom
[John-ga
John-nom
t katta
bought
koto]-ga
fact-nom
mondai-da
problem-cop
to
COMP
omotteru
think
no
?Which book is it that Mary thinks the fact that John bought it is a
problem?
LS assign a ??-status to these cases because both examples involve complex NP-
island violations, as the wh-element is scrambled across a clause headed by the
noun koto (?the fact that?) and hence have a nominal status. Whatever degradation
this incurs, so their argument goes, it remains constant for subjects and objects.
In other words, there is no further cost for extraction out of subjects, and hence
subjects are not islands in Japanese.
LS?s data reveals a quite general complication with the investigation of subex-
traction in Japanese. The only way to construct a scenario where subextraction can
be tested involves complex NPs, headed either by koto or no.1 Japanese does not
allow for bare finite or infinite sentential subjects. PP-subextraction out of NPs,
split-constructions or quantifier floating are unavailable as well.
However, there are two other confounding factors that could have been con-
trolled for. First, the examples in (3) differ in that the object clause involves one
more level of embedding compared to the subject. In (3-b) the matrix clause ?Mary
thinks? is added, which is absent in (3-a). LS are forced to do so to exclude a parsing
1No, which is used in (1), also has a nominal status. It is often referred to as a ?nominalizing
complementizer? although it is translated as ?that? in this example. It is unclear why this example
is not as degraded as the two examples in (3), but it does not seem to be the case that no does
not induce the CNPC effects because replacing koto with no does not seem to make the sentence
any better than ??, i.e. both induce CNPC islands.
179
of the string in (3-a) where the movement of ?which book? is construed as clause-
internal scrambling, which would not involve subextraction. This is illustrated by
(4):
(4) dono
which
hon-o
book-acc
[mary-ga
mary-nom
t kat-ta
bought
koto]-ga
fact-nom
mondai
problem
na
be
no?
Q
?which book is [that mary bought t] a problem??
This string could also be interpreted as scrambling within the koto-clause, as in (5).
In that case, there would be no scrambling ?out of? a subject.
(5) [which book [mary-nom bought t fact]]-nom problem be?
Adding the extra clause blocks this interpretation and guarantees that the wh-
phrase has undergone long-distance scrambling. The same is not necessary in (3-b),
as the fact that ?which book? is to the left of ?Mary? unambiguously indicates that
long-distance scrambling has taken place.
Second, LS do not consider the status of the baseline conditions, i.e. whether
there is a contrast between sentential subjects vs. sentential objects without scram-
bling. This is particularly relevant in the case at hand, as the object but not the
subject example involves one level of center-embedding, which is notorious in incur-
ring a higher processing cost and consequently a lower acceptability.
To sum up, a quick survey of the theoretical literature on subject islands in
Japanese leaves a number of questions. The fact that extraction out of subjects is
less degraded than extraction out of adjuncts does not imply that subject islands do
not exist in Japanese. Adirect comparison between subject andobject subextraction
at first glance suggests that there is no contrast. However, a number of potentially
confounding factors have not been controlled for. In addition, many of the pertinent
judgments are very subtle, and prone to a considerable amount of inter-speaker
180
variation. In short, the empirical evidence on the absence of the subject condition
in Japanese is murky, a fact which calls for a more careful look at the data.
6.1.2 Experiment 12 - Scrambling and Clefting
6.1.2.1 Design and Methodology
We took LS?s cases as the point of departure but modified and supplemented them in
a number of ways. The study had a 3x2 design, manipulating the factors Sub/Obj
and Scrambling and Clefting. This yields the following 6 conditions:
(6) a. Subject, -scrambling
sono
the
syouzyo-wa
girl-Top
iziwaruna
mean
ane-ga
sister-Nom
kuma-no-nuigurumi-o
teddy.bear-Acc
suteta
dumped
koto-ga
fact-Nom
kenka-no-genin
fight-gen-cause
da
Cop
to
Comp
uttaeta
claimed
?The girl claimed that the fact that the mean sister dumped her teddy
bear is the cause of the fight.?
b. Subject, +scrambling
kuma-no-nuigurumi-o
teddy.bear-Acc
sono
the
syouzyo-wa
girl-Top
iziwaruna
mean
ane-ga
sister-Nom
suteta
dumped
koto-ga
fact-Nom
kenka-no-genin
fight-gen-cause
da
Cop
to
Comp
uttaeta
claimed
Her teddy bear1, the girl claimed [that [the fact that the mean sister
dumped ti] is the cause of the fight].
c. Object, -scrambling
sono
the
syouzyo-wa
girl-Top
iziwaruna
mean
ane-ga
sister-Nom
kuma-no-nuigurumi-o
teddy.bear-Acc
suteta
dumped
koto-o
fact-Acc
naisyo-ni-siteita
secret-Dat-kept
to
Comp
uttaeta
claimed
?The girl claimed that her sister kept as a secret the fact that she dumped
her teddy bear.?
d. Object, +scrambling
kuma-no-nuigurumi-o
teddy.bear-Acc
sono
the
syouzyo-wa
girl-Top
iziwaruna
mean
ane-ga
sister-Nom
suteta
dumped
koto-o
fact-Acc
naisyo-ni-siteita
secret-Dat-kept
to
Comp
uttaeta
claimed
181
Her teddy bear1, the girl claimed [that her sister kept as a secret [the
fact that she dumped t1]]
e. Clefting from within a subject clause
Sono
The
syouzyo-ga
girl-Nom
iziwaruna
mean
ane-ga
sister-Nom
e1 suteta
dumped
koto-ga
fact-Nom
kenka-no
fight-Gen
genin
cause
da
Cop
to
Comp
uttaeta
claimed
no-wa
NL-Top
kuma-no-nuigurumi1
teddy-bear
da
Cop
?It is her teddy bear1 that the girl claimed that the fact that her mean
sister dumped e1 is the cause of the fight.?
f. Clefting from within an object clause
Sono
The
syouzyo-ga
girl-Nom
iziwaruna
mean
ane-ga2
sister-Nom
PRO2 e1 suteta
dumped
koto-o
fact-Acc
naisyo-ni-siteita
secret-Dat-kept
to
Comp
uttaeta
claimed
no-wa
NL-Top
kuma-no-nuigurumi1
teddy-bear
da.
Cop
?It is her teddy bear1 that the girl claimed that her sister kept as a
secret the fact that she dumped e1.?
We had 18 sets of items (3 tokes per condition for each subject) which were grouped
into 6 Latin-square lists. A sub-experiment with same structure (18 sentence sets)
as well as 24 filler items were added. Every participant saw a total of 60 items.
27 native speakers of Japanese (by self assessment) without any prior training in
linguistics participated in the study. The study was conducted online using Alex
Drummond?s spellout software.
The number of clauses was controlled for. We followed LS in adding a third
clause to the subject condition example to guarantee that long-distance scrambling
took place, but we counterbalanced this by adding a third clause in the object
condition as well. This, however, resulted in a sequence of three subjects in the
object conditions in (6-c) and (6-d). Since three subjects in a row come close to
being unparsable, we replaced the lowest subject with a PRO, which is controlled
by the second lowest subject (see Fujii 2006, among others, for control phenomena
across a finite clause in Japanese). This eases the parsability of the string in the
object condition and thus helps to keep the number of clauses constant. On the other
182
hand, this creates a garden path, as the parser only knows that there is a PRO when
it reaches the end of the clause. This constitutes an asymmetry compared to the
subject conditions in (6-a) and (6-b), which is likely to reflect in the acceptability
rating. We will return to this issue in the discussion section below.2 Finally, the
baseline conditions without scrambling were added in (6-a) and (6-c).
6.1.2.2 Results
The results confirmed LS?s intuitions that there is no significant difference between
scrambling out of subjects vs. objects, and no difference between operator move-
ment from subjects vs. objects in clefts. We do find, however, a significant difference
between the baseline conditions, with subjects being preferred over objects. Cru-
cially, there is also a significant interaction effect between the factors Sub/Obj and
Extraction. The results are summarized in Figure 6.1 and Tables 6.1 and 6.2.
Pair-wise comparisons confirmed LS?s intuitions that there is no significant
difference between scrambling out of subjects (2.73) vs. objects (2.85), and no
difference between operator movement from subjects (2.57) vs. objects (2.96) in
clefts (t<.1). We do find, however, a significant difference between the baseline
conditions, with subjects (6.93) being preferred over objects (5.79) (t(1,80)=2.9,
p=.002).
Table 6.1: Japanese: means
Mean SD
sub, -ext 6.39 .96
sub, +scr 2.73 1.51
sub, +cleft 2.57 1.24
obj, -ext 5.79 1.63
obj, +scr 2.85 1.59
obj, +cleft 2.96 1.33
Table 6.2: ANOVA Japanese
F p
Sub/Obj 24.945 ***
Extraction 733.581 ***
Sub/Obj*Extraction 61.366 ***
2Another minor difference between our examples and LS?s is that we changed long-distance
wh-scrambling to long-distance non-wh scrambling.
183
gid23gid1
gid24gid1
gid25gid1
gid26gid1
gid27gid1
gid28gid1
gid29gid1
gid21gid7gid19gid15gid13gid4gid6gid16gid12gid11gid1 gid30gid14gid6gid13gid1 gid30gid6gid10gid7gid8gid1
gid3gid18gid5gid9gid7gid6gid15gid1
gid2gid5gid9gid7gid6gid15gid1
Figure 6.1: Extraction out of non-finite clauses in Japanese
6.1.2.3 Discussion
The lack of a significant difference between scrambling out of subjects vs. objects
might tempt us to follow LS in concluding that the subject condition does not
hold in Japanese. However, once we take the baseline conditions into consideration
a different picture emerges. The object baseline condition, (6-c), is significantly
degraded compared to the subject condition, (6-a). Extraction disproportionally
affects subjects worse than objects - the Sub/Obj*Extraction interaction effect
in Table 6.2. While long-distance scrambling leads to strong degradation in both
case due to the unavoidable CNPC effect induced by koto, subjects are more strongly
affected than objects. We take this to be highly suggestive of the existence of the
Subject Condition in Japanese.
What are the origins of the asymmetry between the baseline conditions? We
have already hinted above at the potential source. Even though the number of
clauses inbothconditions isthesame, the object conditionsentences aresignificantly
184
more complex than the subject condition. Only the object condition involves center-
embedding. This is obviously a disadvantage for the object condition. The object
condition also involves three nominal subjects (?girl?, ?sister?, and PRO controlled
by ?sister?). The subject condition involves two nominal subjects: ?girl? and ?sister?.
The calculation of the controller of PRO may be an extra burden in the object
condition. The parser only realizes that the sentence contains a PRO once it reaches
?kept? at the very end of the sentence. The alternative to the PRO would have been
3 overt subjects in a row. However, we will see from filler-data below that center-
embedding with 3 overt subjects even when no extraction takes place leads to worse
acceptability than either of our scrambling conditions. This suggest that replacing
one subject with a PRO helped the object condition to a certain extent but still did
not boost it all the way to the status of the subject condition.
Given that the ratings for both scrambling conditions are relatively low on the
scale, it is a valid concern that we have produced floor effects and that potential
differences have been washed out. Looking at our filler data, however, we can be
fairly confident that our conditions are not at the floor of the scale. Table 6.2 gives
the rating of some representative filler items in (7) and compares them with the two
scrambling conditions.
(7) a. Long-distance scrambling
Igirisusei-no
British
koutya-o1
tea-Acc
masutaa-wa
cafe-owner-Top
sono
the
tyuunen-no
middle-aged-Gen
sinsi-ga
gentlemen-Nom
t1 maikai
every-time
oodaasuru
order
to
Comp
hanasi-tei-ta.
talk-prog-past
?British tea1, the coffee shop owner was talking that the middle aged
gentlemen orders t1 every time.?
b. Center-embedding
Obasan-ga
Aunt-Nom
syouzikina
honest
[bebiisittaa-ga
baby-sitter-Nom
[ani-ga
older-brother-Nom
imouto-o
younger-sister-Acc
izimeta
bullied
to]
Comp
itta
said
to]
Comp
omot-tei-ru.
think-prog-pres
185
?The aunt thinks [that the honest babysitter said [that the older brother
bullied his younger sister]].?
c. Why in-situ inside RC
Hudan-wa
Usual-Top
mazimena
diligent
sono
the
gakusei-wa
student-Top
[amerikazin-no
[American-Gen
eigo-no
English-Gen
sensei-ga
teacher-Nom
naze
why
dasita
assigned
syukudai]-o
homework]-Acc
yari-tagara-nakat-ta
do-want-not-past
no?
Q
?The usually-diligent student didn?t want to do [the homework that the
American English teacher assigned why]??
d. CSC violation
Amaku-te
Sweet-and
oisii
delicious
ringo-to1
apple-and
kuisinbouno
gluttonous
syougakusei-ga
grade-schooler-Nom
t1
suppai
sour
mikan-o
orange-Acc
douzi-ni
at-the-same-time
kuti-ni
mouth-to
ireta.
put
?A sweet and delicious apple1, the gluttonous grade schooler put [t1 and
a sour orange] into his mouth at the same time.?
gid33gid1
gid34gid1
gid35gid1
gid36gid1
gid37gid1
gid38gid1
gid39gid1
gid1gid15gid18gid17gid11gid30gid9gid13gid20gid21gid6gid17gid8gid10gid1
gid20gid8gid19gid6gid16gid7gid15gid13gid17gid11gid1
gid5gid24gid7gid14gid29gid1gid40gid20gid8gid19gid1 gid3gid7gid14gid29gid1gid30gid20gid8gid19gid1 gid2gid10gid17gid21gid10gid19gid30gid10gid16gid7gid10gid9gid9gid13gid17gid11gid1 gid26gid12gid27gid31gid1gid13gid17gid30gid20gid13gid21gid24gid1gid13gid17gid20gid13gid9gid10gid1
gid4gid2gid1
gid2gid5gid2gid1gid25gid13gid18gid15gid6gid22gid18gid17gid1
Figure 6.2: Scrambling conditions compared with filler items
We see that grammatical long-distance scrambling already results in slightly de-
creased acceptability. Crucially, center-embedding with three overt subjects, wh-
adjuncts inside a relative clause and CSC-violations are all worse than our target
186
conditions. This strongly suggest that our conditions are still located within the
meaningful range of the scale.
6.1.3 Conclusion
Our study allows usto conclude that subject island effects exist inJapanese, contrary
to the majority view in the theoretical literature. However, it is only fair to point out
that our results are not as conclusive as we would like them to be. LS?s intuition that
extraction out of subject and objects is equally degraded still holds. Our evidence
is more indirect and purely stems from the interaction effect, which is solely due to
the asymmetry in the baseline conditions. We discussed a number of confounding
factors in Japanese that make it hard to construct perfect stimuli. Nonetheless, it
seems fair to conlude that we have shown that the conclusion that the CED does
not hold is too simple. Further studies will have to be conducted to make stronger
claims about the empirical facts in Japanese.
6.2 Serbian3
6.2.1 Introduction
Island effects in Serbian have received far less attention than locality in German,
English or Japanese (even though some intriguing work has been done). This short
section will not present a full-fledged theory of locality effects in Serbian or even
the status of the CED, but we will present two studies that show a surprising
asymmetry. Concretely, acceptability judgment data indicates that PP-extraction
(PPE) in Serbian shows the familiar subject/object asymmetries, while there are no
such effects for Left Branch Extraction (LBE). This might very well be evidence for
an account pursued by ?Cavar & Fanselow 2002 and Ba?si?c 2004, who argue that LBE
3The experiments discussed are joint work with Ivana Mitrovi?c
187
is best analysed as an instance of scattered deletion. If PPE involves subextraction
of the kind observed in German and English, while LBE does not, CED effects would
be expected for the former but not the latter. It also a possibility that what we call
PPE in Serbian is in fact a hanging topic construction, similar to what we argued
for in the English pied-piping case in section 5.2.
It is well-known that a number of Slavic languages allow LBE, while languages
like English do not (Serbian example from B?asi?c 2004).
(8) a. *Whose did they close factory?
b. Koju
which
su
aux
zatvorili
closed
fabriku?
factory
?Which factory did they close??
It has also been noted that LBE in Serbian does not violate the Subject condition.
Both examples in (9) indicate that the left branch can originate in a preverbal
subject position:
(9) a. Ovaj
this
nam
us-cl
je
aux-cl
predsednik
president
obe?cao
promised
ve?ce
higher
plate.
salaries
?This president promised us higher salaries.?
b. ?Ciju
whose
je
aux
on
he
knjigu
book
preveo?
translated
?Whose book did he translate??
[Ba?si?c 2004, ex. (61a) and ex. (64c)]
While Serbian does not allow preposition stranding, fronting a full PP is licit:
(10) a. *Kome
whom
su
aux
one
they
objavile
publish
?clanak
an.article
o?
about
?Who did they publish an article about??
b. O
about
kome
whom
su
aux
one
they
objavile
publish
?clanak?
an.article
?Who did they publish an article about??
188
Is LBE really completely insensitive to subject/object asymmetries? And is there
a strong difference compared to postnominal subextraction? Ba?si?c (2004, 34-36)
concludes that ?the extraction of postnominal elements and prenominal elements
seems not to be subject to the same conditions?. She also concedes, however, that
?judgements concerning the grammaticality of extraction from DPs [...] seem to
vary considerably?.
The purpose of our study is to complement the empirical discussion of sub-
ject/object subextraction asymmetries in Serbian, a language in which controlled
acceptability judgment studies have rarely been conducted.4
6.2.2 Experiment 13: Left Branch Extraction vs. PP-extraction
6.2.2.1 Design and Methodology
This experiment consists of two subexperiments. We look at subject/object asym-
metries for Left Branch Extraction (LBE), on the one hand, and PP-extraction
(PPE) on the other hand. For both constructions we manipulated the factors Sub-
Obj and Extraction, which yields the following 4 conditions each:
4The fact that subextraction phenomena in Serbian could strongly benefit from more rigid
methodology is illustrated by the following example from Bo?skovi?c 2005, p. 32:
i ?Cimei
what.instr
ga
him
je
is
[pretnja
threat
ti]
scared
upla?sila?
?The threat of what scared him??
Many speakers strongly disagree with Bo?skovi?c?s judgments according to which subject subextrac-
tion is acceptable in Serbian and find (i) severely degraded. We added this sentence as filler item
to our study below, where it was given an intermediate judgment of 3.45 with a relatively high
standard deviation of 1.91. What the source of the disagreement about this example is will not be
settled here. It is important to note, however, that its status is controversial.
189
(11) Left branch extraction
a. Subject, -extraction
?Ciji
Whose
(10
friends
prijatelji
AUX
su
you
te
yesterday
ju?ce
after
posle
noon
podne
meet?
upoznali?
?Whose friends met you yesterday in the afternoon??
b. Subject, +extraction
?Ciji
Whose
su
AUX
te
you
ju?ce
yesterday
posle
after
podne
noon
prijatelji
friends
upoznali?
meet?
?Whose friends met you yesterday in the afternoon??
c. Object, -extraction
?Cije
Whose
prijatelje
friends
si
AUX
ju?ce
yesterday
posle
after
podne
noon
upoznao?
meet?
?Whose friends did you meet yesterday in the afternoon??
d. Object, +extraction
?Cije
Whose
si
AUX
ju?ce
yesterday
posle
after
podne
noon
upoznao
meet
prijatelje?
friends?
?Whose friends did you meet yesterday in the afternoon??
(12) PP-extraction
a. Subject, -extraction
Knjiga
Books
o
about
Marku
Markus
je
AUX
pro?sle
last
godine
summer
izazvala
caused
burne
turbulent
polemike.
controversy
?Books about Markus caused a controversy last summer.?
b. Subject, +extraction
O
about
kome
whom
je
AUX
pro?sle
last
godine
summer
knjiga
books
izazvala
caused
burne
turbulent
polemike?
controversy
?About whom did books cause a controversy last summer??
c. Object, -extraction
Taj
that
politi?car
politician
je
AUX
pro?sle
last
godine
summer
pro?citao
read
knjigu
books
o
about
Marku.
Markus
The politician read books about Markus last summer.?
d. Object, +extraction
O
about
kome
whom
je
AUX
pro?sle
last
godine
summer
taj
that
politi?car
politician
pro?citao
read
knjigu?
books
?About whom did the politician read books last summer??
190
Three lexicalizations of each condition were constructed and grouped into four Latin-
square lists (two separate sets of lists for PPE and LBE). Each participant was
presented 24 conditions and 36 fillers items of all levels of acceptability. 20 native
speakers of Serbian with no prior training in linguistics participated in the study
online.
6.2.2.2 Results
Interestingly, Serbian shows subject/object asymmetries withPP-extraction (?x=3.22
vs. 5.50, t(1,59)=6.75, p=<.001) but not with LBE (?x=3.35 vs. 3.02, t(1,59)=.92,
p=.18), as illustrated in Figure 6.3 2. The black lines represent the object condi-
tions and the grey lines the subject conditions, the solid lines refer to the PPE and
the dotted lines to the LBE conditions. It is immediately noticeable that the solid
lines are almost perfectly parallel while the dotted lines show are non-parallel, indi-
cating an interaction effect between the two factor. Two 2 way repeated measures
ANOVAs confirm that there is a significant SubObj and Extraction for PPE
(F(1,59)=170.031, p<.001) but not for LBE (F(1,59)=.104, p<.749).
gid19gid15gid18gid18gid1
gid20gid15gid18gid18gid1
gid21gid15gid18gid18gid1
gid22gid15gid18gid18gid1
gid23gid15gid18gid18gid1
gid24gid15gid18gid18gid1
gid25gid15gid18gid18gid1
gid16gid7gid13gid11gid1 gid26gid7gid13gid11gid1
gid4gid2gid3gid14gid1gid10gid12gid6gid1
gid4gid2gid3gid14gid1gid9gid6gid8gid1
gid5gid5gid3gid14gid1gid10gid12gid6gid1
gid5gid5gid3gid14gid1gid9gid6gid8gid1
Figure 6.3: Left branch extraction and PP-extraction in Serbian
191
6.2.2.3 Discussion
Our results complement the intuitions from the theoretical literature and suggest
that LBE is not sensitive to subject islands while PPE is. This prompted a number
of authors to conclude that LBE is not a case of genuine subextraction: Ba?si?c (2004)
proposes an analysis in terms of remnant movement and ?Cavar & Fanselow 2002 as
well as Pereltsvaig (2006) - basing her argument on Colloquial Russian - advocate
for a scattered deletion account. Other authors such as Bo?skovi?c (2005) or Corver
(1992) defend a more traditional analysis in terms of direct extraction.
It is possible that the PPE cases do not involve genuine subextraction either
but are hanging topic constructions, similar to what we argued for pied-piping in
English insection 5.2. It isnot thepurpose ofthissection to settle thistheoretical de-
bate. It seems fair to conclude, however, that any theory of LBE, PP-subextraction
and the CED in Serbian most take results of the sort we presented into account.
Furthermore, we hope that future research in the syntax of Serbian will employ
acceptability judgment studies as a supplement to the theoretical discussion.
192
Chapter 7
Conclusion
About 200 pages and 13 experiments later it is time to reflect on what we have
learned from this dissertation:
? We have strong reasons to believe that the CED remains a empirically valid
generalization. Our studies all converge on the conclusion that complements
are the preferred domain of extraction.
? Linguistics judgments are not matters of personal taste that are not amenable
to scientific inquiry, but they can be subjected to experimental scrutiny. The
reader is hopefully convinced that acceptability judgment studies are a useful
tool when we are faced with controversial judgments. We have emphasized,
however, that they should by no means be regarded as a replacement of theo-
retical work but as a helpful supplement.
? There isstrong evidence thatGermanNP-subextraction shows complement/non-
complement asymmetries that are independent of freezing effects. Freezing is
an additional factor that further decreases the acceptability. In other words,
we have provided a diagnostic for base positions, i.e., we can make a three-way
distinction between complements, specifiers, and derived specifiers.
? Theoretically, this seems to indicate that not one but two theories are needed:
an account of the CED (possibly in terms of MSO) as well as an explanations
of freezing effects.
193
? We have provided further evidence that the LCA in its most orthodox inter-
pretation cannot be maintained for SOV languages like German.
? We saw that indirect objects are opaque domains, at least in German. This
is expected under a CED-view of the world if they are specifiers or adjuncts.
The situation becomes much more mysterious if indirect objects are analysed
as complements (as in Baker 1988 or Larson 1988).
? When we investigate subextraction phenomena, we have to make sure that
the phenomenon we are investigating in fact involves subextraction. Pied-
piping in English, for instance, misled us into believing that English subjects
are sometimes transparent, while in fact it is highly plausible that no genuine
extraction has taken place. Left Branch Extraction in Serbian is likely to be
a similar case.
? ATB-was-f?ur split can ameliorate island violations in way comparable to PG-
constructions. We have arguedfor ananalysis in terms of sidewards movement.
There are many issues we covered but many more we just touched upon, which
leaves ample space for future research. Perhaps this dissertation will prove useful
when the following issues are addressed in future work:
? What is the relation between grammaticality and acceptability? How big is
the role that extra-grammatical factors such as processing and information
structure play in general, and in the domain of islands in particular?
? A related question about the nature of our grammar arises. Is it gradient? Or
is it binary and all the gradience we find is due to extra-grammatical factors?
What does it mean for a sentence to violate a grammatical constraint?
? There are other languages that are sometimes claimed to violate the CED
194
(e.g. Turkish, Russian). Do these cases hold up in controlled acceptability
judgment studies?
? The domain of islands in general is full of controversial judgments. What about
adjunct islands cross-linguistically? Wh-islands? Counterexamples have been
put forth to virtually all island constraints. Do they in fact hold as reported,
and can acceptability judgment studies help us complete the picture?
? A more detailed empirical landscape of island constraints can help us answer
the question of whether what we refer to as island phenomena even constitutes
a coherent group. Perhaps some islands require syntactic or semantic explana-
tions, others might be more amenable to processing or information structural
accounts.
.
195
Appendix A
Instructions
A.1 German
Sprachwissenschaftliche Studie
Einverst?andniserkl?arung
Diese Studie untersucht, wie MuttersprachlerInnen des Deutschen unterschiedliche
S?atze in ihrer Sprache beurteilen. Die Informationen, die in dieser Studie gesam-
melt werden, werden streng vertraulich behandelt. Die Daten werden anonym und
gemeinsam mit denen anderer Testpersonen statistisch ausgewertet. Mein Name
scheint in der Datenanalyse NICHT auf.
Ich habe verstanden, dass dieses Experiment NICHT meiner pers?onlichen
Weiterbildung dient. Die Studie dient dem Zweck, Erkenntnisse ?uber die Beschaf-
fenheit menschlicher Sprache zu gewinnen. Ich habe jederzeit die M?oglichkeit das
Experiment ohne Nennung eines Grundes abzubrechen.
Fragen und Kommentare bitte an:
Johannes Jurka
University of Maryland, Department of Linguistics
jjurka@umd.edu
Ich best?atige, dass ich ?uber achtzehn Jahrealt bin, dass ichobenangef?uhrteErl?auterun-
gen aufmerksam gelesen habe und dass ich mich bereit erkl?are, an dieser sprachwis-
senschaftlichen Studie teilzunehmen.
Name und Datum: Unterschrift:
Graduelle Akzeptabilit?atsurteile
I. Indieser Studiewerden Sieersucht, S?atzedes Deutschen IhremSprachgef?uhl
als MuttersprachlerIn folgend auf einer Skala von 1-7 einzustufen. Bitte beurteilen
Sie die S?atze nach ihrer Akzeptabilit?at, d.h. ob sie Ihnen intuitiv als nat?urliche
S?atze des Deutschen vorkommen.
? Wenn Sie glauben, dass der Satz ein einwandfreier Satz des Deutschen ist,
geben Sie ihm eine hohe Bewertung (6 oder 7).
196
? Wenn Sie glauben, dass der Satz kein m?oglicher Satz des Deutschen ist, dann
geben Sie ihm eine niedrige Bewertung (1 oder 2).
? Manche S?atze kommen Ihnen vielleicht nicht komplett inakzeptabel vor, klin-
gen aber auch nicht ganz perfekt - geben Sie solchen S?atzen eine mittelm?a?ige
Bewertung (3-5).
II. Wichtig ist, dass es NICHT dass es NICHT um die Plausibilit?t eines Satzes
geht. Bitte beurteilen Sie die S?atze nur danach, ob sie sich f?ur Sie als m?ogliche S?atze
des Deutschen anh?oren oder nicht. Folgende Beispiele illustrieren diesen Punkt:
(1) Die Kinder haben den Weihnachtsbaum mit Girlanden geschm?uckt.
(schlecht) 1 2 3 4 5 6 7 (gut)
(2) Die Kinder haben Girlanden auf den Weihnachtsbaum geschm?uckt.
(schlecht) 1 2 3 4 5 6 7 (gut)
(3) Das blassblaue Nilpferd hat mit dem H?angebauchschwein Schach gespielt.
(schlecht) 1 2 3 4 5 6 7 (gut)
Beispiel (2) beschreibt eine durchaus plausible Situation. Dennoch empfinden die
meisten DeutschsprecherInnen diesen Satz als inakzeptabel (im Gegensatz zu (1),
der f?ur die meisten SprecherInnen einwandfrei ist).
Beispiel (3), andererseits, beschreibt eine relativ seltsame Situation. Dennoch
w?are es problemlos m?oglich so ein Szenario (etwa in einem Zeichentrickfilm) mit
diesem Satz zu beschreiben.
III. Au?erdem ist es wichtig zu betonen, dass es NICHT darum geht zu
beurteilen, ob ein Satz den Regeln des Dudens oder der Schulgrammatik entspricht
bzw. ob der Satz in einem formellen Kontext verwendet wird. Bei der Beurteilung
geht es lediglich darum, zu bewerten, ob sich ein Satz intuitiv nat?urlich anh?ort
und in einer angemessenen Situation von einer SprecherIn des Deutschen verwendet
werden k?onnte oder nicht.
ZumBeispiel k?onnte ineinem Schulaufsatz ausgebessert werden, dass ?w?ahrend?
den zweiten Fall verlangt, z.B. ?w?ahrend des Essens?. Die meisten SprecherIn-
nen des Deutschen w?urden jedoch umgangssprachlich folgenden Satz von Ihrem
Sprachgef?uhl her als einwandfreien und nat?urlichen Satz des Deutschen einstufen.
Des Weiteren spielen Rechtschreibung (egal, ob alte oder neue) und Beistrich-
setzung KEINE Rolle f?ur die Beurteilung der S?atze in diesem Experiment.
(4) Ich habe w?ahrend dem Essen mit meiner Freundin telefoniert..
(schlecht) 1 2 3 4 5 6 7 (gut)
197
Des Weiteren spielen Rechtschreibung (egal, ob alte oder neue) und Beistrichsetzung
KEINE Rolle f?ur die Beurteilung der S?atze in diesem Experiment.
Bevor wir beginnen, hier sind einige Beispiele f?ur m?ogliche Bewertungen:
(5) Welchen Rum hat denn der Kellner das Cola und in ein Glas geleert?
(schlecht) 1 2 3 4 5 6 7 (gut)
(6) Welchen Politiker hast du denn gestern vor dem Rathaus getroffen?
(schlecht) 1 2 3 4 5 6 7 (gut)
(7) Welches Buch fragst du dich denn oft, ob der Student gelesen hat?
(schlecht) 1 2 3 4 5 6 7 (gut)
Ich w?urde Sie zuletzt noch bitten, die S?atze m?oglichst schnell und Ihrer ersten
Intuition folgend zu beurteilen. Sobald Sie bereit sind, kann es auf der n?achsten
Seite losgehen!
Vielen Dank!
198
A.2 English
I hereby confirm that I am over 18 years of age and that I am a native speaker of
English. I am aware that I am participating in a linguistic experiment. This study
is fully anonymous and the data will only be used for statistical analysis. I can
abort the experiment at any time.
1. In the following study you will be ask to rate sentences of English according
to your intuitive judgments as a native speaker. The experiment will take between
10-15 minutes of your time. If you have any questions or comments please email:
jjurka@umd.edu.
Sentence Acceptability Ratings
For the list of sentences below, please rate whether each sentence seems like an
acceptable sentence in everyday English.
? If you think that the sentence sounds acceptable and possible in English, then
you should give it a high rating (6 or 7)
? If you think that the sentence does not sound like a possible sentence of En-
glish, then you should give it a low rating (1 or 2).
? Some sentences may not sound like totally impossible sentences, but are also
not completely acceptable ? you could give those a more intermediate rating
(3-5).
2. You are NOT being asked to judge the plausibility of the meaning of the
sentence; you are simply being asked to judge whether the sentence sounds like
possible English or not. Consider the following examples:
Example (9)below describes ahighly likely scenario, but most English speakers
find it unacceptable (unlike a similar sentence in (8)), and could not use it.
(8) The children decorated the tree with sparkling ornaments.
(bad) 1 2 3 4 5 6 7 (good)
(9) The children decorated the sparkling ornaments onto the tree.
(bad) 1 2 3 4 5 6 7 (good)
(10) The purple elephant played chess with the balding porcupines.
(bad) 1 2 3 4 5 6 7 (good)
In contrast, example (10) describes an implausible and outlandish situation, but if
it were necessary to describe such a crazy scenario, you could use the sentence in
(10) without any problem.
199
3. You are also NOT being asked to judge whether the sentence is acceptable
according to ?school grammar?, i.e., the rules you may have learned in high school
English, or in writing classes. You?re just being asked to judge whether the sentence
sounds like natural English that you or other speakers of English might be able to
use.
For example, you might have learned not to end a sentence with a preposition.
However, while the example (d) ends with a preposition, most English speakers find
it completely natural, acceptable sentences.
(11) The old woman hated the people who she was traveling with.
(bad) 1 2 3 4 5 6 7 (good)
Here are some more example sentences and ratings before you move on to the test:
(12) Jane threw out the lemon that Rick squirted the lime and in his drink.
(bad) 1 2 3 4 5 6 7 (good)
(13) I ate the chili that Mary left out on the table.
(bad) 1 2 3 4 5 6 7 (good)
(14) Which book do you wonder whether James read yesterday?
(bad) 1 2 3 4 5 6 7 (good)
Please give your ratings by either clicking the number on the screen or pressing the
button on your keyboard. Thank you!
200
A.3 Serbian
Da biste u?cestvovali u ovom ispitivanju, morate imati vi?se od 18 godina i morate biti
izvorni govornik srpskog jezika. Ovo je lingvisti?cki eksperiment koji je u potpunosti
anoniman. Rezultati ispitivanja bi?ce upotrebljeni u svrhe statisti?cke analize. Mo?zete
odustati od ovog istra?zivanja u svakom momentu bez ikakvog obja?snjenja. Ukoliko
imate bilo kakva pitanja u vezi s ovim istra?zivanjem, molim Vas da kontaktirate
Ivanu Mitrovi?c na imejl adresu: hellasbb@gmail.com.
Rangiranje prihvatljivosti re?cenica
1. Molim Vas da dole navedene re?cenice rangirate po prihvatljivosti u svakodnevnom
srpskom jeziku.
? Ako mislite da je re?cenica u potpunosti prihvatljiva i mogu?ca u srpskom jeziku,
ocenite je visoko (6 ili 7 poena).
? Ako mislite da re?cenica ne zvu?ci kao mogu?ca re?cenica srpskog jezika, dajte joj
nisku ocenu (1 ili 2 poena).
? Neke od re?cenica mogu zvu?cati kao ne ba?s potpuno neprihvatljive a u isto
vreme i ne ba?s potpuno prihvatljive. U tom slu?caju ih ocenite prose?cnom
ocenom (3-5 poena).
2. Ono ?sto NE treba da ocenjujete je zna?cenje re?cenice. Samo treba da ocenite
da li re?cenica zvu?ci kao mogu?ca re?cenica u srpskom jeziku bez obzira na njeno
zna?cenje. Na primer, re?cenica u primeru (b) opisuje krajnje mogu?cu situaciju, ali
je ve?cina govornika srpskog jezika ocenjuje kao neprihvatljivu i ne bi je upotrebili
(za razliku od sli?cne re?cenice u primeru (a)). Re?cenica u primeru (c) opisuje veoma
?cudnu situaciju, ali ukoliko je potrebno opisati ba?s takvu situaciju, sasvim je mogu?ce
upotrebiti re?cenicu u primeru (c).
a Okitili smo jelku svetle?cim ukrasima.
(lo?se) 1 2 3 4 5 6 7 (dobro)
b Okitili smo svetle?ce ukrase jelkom.
(lo?se) 1 2 3 4 5 6 7 (dobro)
c Ljubi?casti slon je igrao ?sah s golobradim je?zevima.
(lo?se) 1 2 3 4 5 6 7 (dobro)
3. Takodje, NE treba da ocenjujete prihvatljivost re?cenica po gramati?ckim pravilima
koje ste u?cili u?skoli; to jest, po pravilima koja su napisana u gramatikama. Treba da
ocenite prihvatljivost re?cenica po tome da li Vama one zvu?ce kao mogu?ce re?cenice u
srpskom jeziku koje biste Vi ili neki drugi govornici srpskog jezika mogli upotrebiti.
201
Ono ?sto nas zanima je ?sta Vama zvu?ci kao dobra re?cenica u srpskom a ne ?sta ka?zu
gramati?cka pravila.
Na primer, u ?skoli ste nau?cili gramati?cko pravilo po kome oblik glagola biti
za prvo lice mno?zine glasi bismo. Iako ovo pravilo postoji u gramatici, veliki broj
govornika srpskog jezika koristi oblik bi. Gramatike ocenjuje ovaj oblik kao neprav-
ilan, dok ga izvorni govornici srpskog jezika upotrebljavaju i ne ocenjuju ga kao
nepravilan (primer (d)).
d Mi bi i?sli u grad.
(lo?se) 1 2 3 4 5 6 7 (dobro)
4. Neke re?cenice u srpskom jeziku su u potpunosti prihvatljive iako su poprili?cno
duga?cke i slo?zene. Re?cenica prikazana u primeru (e) mo?ze izgledati na prvi pogled
pomalo duga?cka i neshvatljiva ali je u potpunosti prihvatljiva (iako podu?za). Dakle,
ocenili biste je visokom ocenom 6 ili 7 poena. Za razliku od nje, re?cenica u primeru
(b) je kratka ali je potpuno neprihvatljiva.
e Dinki?c je rekao da je sa MMF-om na samom po?cetku vi?sednevnih razgovora
dogovoreno da deficit bud?zeta bude ?cetri odsto bruto doma?ceg proizvoda, a
ne 3,5 odsto koliko je MMF ranije tra?zio i dodao da ne?ce biti problema oko
bud?zeta za 2010, ali da ostaje izazov kako na srednji rok odr?zati ravnote?zu,
kao i to da su prihodi bud?zeta bolji nego pred prethodni dolazak misije.
Evo jo?s nekoliko primera re?cenica i ocena istih pre nego ?sto po?cnete s istra?zivanjem:
f Jovana je bacila limun koji je Marko iscedio pomorand?zu i u njegovo pi?ce.
(lo?se) 1 2 3 4 5 6 7 (dobro)
g Pojela sam kola?c ?sto je mama ostavila na stolu.
(lo?se) 1 2 3 4 5 6 7 (dobro)
h Koje pismo se pita?s da li je Marko napisao?
(lo?se) 1 2 3 4 5 6 7 (dobro)
Eksperiment: Molimo Vas da zaokru?zite jedan od ponudenih brojeva (1-7) za
svaku re?cenicu.
Ovo je kraj istra?zivanja!
Hvala Vam ?sto ste u?cestvovali u njemu!
202
A.4 Japanese
1. uni3053.860uni306E.887uni8ABF.3024uni67FB.2090uni3067.880uni306F.888uni3001.634uni65E5.3284uni672C.3722uni8A9E.1952uni6BCD.3644uni8A9E.1952uni8A71.4073uni8005.2304uni304C.853uni3069.882uni306E.887uni3088.913uni3046.847uni306A.883uni6587.3592uni3092.923uni81EA.2263uni7136.2741uni3068.881uni611F.1522uni3058.865uni308B.916uni304B.852uni3069.882uni3046.847uni304B.852uni3092.923uni8ABF.3024uni3079.898uni3066.879uni3044.845uni307E.903uni3059.866uni3002.635
uni3053.860uni308C.917uni304B.852uni3089.914uni8AAD.3233uni307E.903uni308C.917uni308B.916uni6587.3592uni306B.884uni5BFE.2864uni3057.864uni3066.879uni3001.634uni76F4.3034uni611F.1522uni7684.3108uni306B.884uni81EA.2263uni7136.2741uni306A.883uni65E5.3284uni672C.3722uni8A9E.1952uni306E.887uni6587.3592uni3067.880uni3042.843uni308B.916uni304B.852uni3069.882uni3046.847uni304B.852uni3092.9231uni304B.852uni3089.9147uni307E.903uni3067.880
uni306E.887uni6570.2618uni5B57.2248uni3092.923uni4F7F.2198uni3063.876uni3066.879uni5224.3409uni65AD.2949uni3057.864uni3066.879uni3044.845uni305F.872uni3060.873uni304D.854uni307E.903uni3059.866uni3002.635
? uni3082.907uni3057.864uni8AAD.3233uni3093.924uni3060.873uni6587.3592uni304C.853uni81EA.2263uni7136.2741uni306A.883uni65E5.3284uni672C.3722uni8A9E.1952uni306E.887uni6587.3592uni3067.880uni3042.843uni308B.916uni3068.881uni611F.1522uni3058.865uni3089.914uni308C.917uni305F.872uni5834.2516uni5408.2041uni306F.888uni3001.6346uni3084.9097uni3068.881uni3044.845uni3063.876uni305F.872uni9AD8.2036uni3044.845uni6570.2618
uni5B57.2248uni3092.923uni9078.2732uni3093.924uni3067.880uni30AF.939uni30EA.998uni30C3.959uni30AF.939uni3057.864uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635
? uni3082.907uni3057.864uni8AAD.3233uni3093.924uni3060.873uni6587.3592uni304C.853uni4E0D.3526uni81EA.2263uni7136.2741uni306A.883uni65E5.3284uni672C.3722uni8A9E.1952uni306E.887uni6587.3592uni3067.880uni3042.843uni308B.916uni3068.881uni611F.1522uni3058.865uni3089.914uni308C.917uni305F.872uni5834.2516uni5408.2041uni306F.888uni3001.6341uni3084.9092uni3068.881uni3044.845uni3063.876uni305F.872uni4F4E.3071uni3044.845
uni6570.2618uni5B57.2248uni3092.923uni30AF.939uni30EA.998uni30C3.959uni30AF.939uni3057.864uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635
? uni3082.907uni3057.864uni8AAD.3233uni3093.924uni3060.873uni6587.3592uni304C.853uni3082.907uni306E.887uni3059.866uni3054.861uni304F.856uni4E0D.3526uni81EA.2263uni7136.2741uni306A.883uni308F.920uni3051.858uni3067.880uni306F.888uni306A.883uni3044.845uni3082.907uni306E.887uni306E.887uni3001.634uni304B.852uni3068.881uni8A00.1908uni3063.876uni3066.879uni81EA.2263uni7136.2741uni306A.883uni6587.3592uni3060.873uni3068.881
uni3082.907uni8A00.1908uni3044.845uni304C.853uni305F.872uni3044.845uni3068.881uni3044.845uni3046.847uni5834.2516uni5408.2041uni306F.888uni3001.6343uni304B.852uni3089.9145uni3068.881uni3044.845uni3063.876uni305F.872uni4E2D.2980uni9593.1554uni306E.887uni6570.2618uni5B57.2248uni3092.923uni30AF.939uni30EA.998uni30C3.959uni30AF.939uni3057.864uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635
2. uni3053.860uni306E.887uni5B9F.2286uni9A13.1895uni3067.880uni8A00.1908uni3046.847uni3068.881uni3053.860uni308D.918uni306E.887uni300C.686uni81EA.2263uni7136.2741uni30FB.638uni4E0D.3526uni81EA.2263uni7136.2741uni300D.687uni306A.883uni6587.3592uni3068.881uni3044.845uni3046.847uni306E.887uni306F.888uni3001.634uni6587.3592uni306B.884uni66F8.2427uni304B.852uni308C.917uni3066.879uni3044.845uni308B.916uni5185.3258uni5BB9.3888uni304C.853uni81EA.2263uni7136.2741
uni306A.883uni72B6.2525uni6CC1.1711uni3092.923uni63CF.3507uni5199.2296uni3057.864uni3066.879uni3044.845uni308B.916uni304B.852uni3001.634uni3068.881uni3044.845uni3046.847uni3053.860uni3068.881uni3067.880uni306F.888uni306A.883uni304F.856uni3001.634uni65E5.3284uni672C.3722uni8A9E.1952uni306E.887uni6587.3592uni3068.881uni3057.864uni3066.879uni53EF.1348uni80FD.3315uni306A.883uni6587.3592uni3067.880uni3042.843uni308B.916uni304B.852uni3069.882uni3046.847
uni304B.852uni304C.853uni554F.3824uni308F.920uni308C.917uni3066.879uni3044.845uni307E.903uni3059.866uni3002.635
uni4F8B.4011uni3048.849uni3070.889uni3001.634uni6B21.2253uni306E.887uni4F8B.4011uni6587.3592uni3092.923uni3054.861uni89A7.3937uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635
? uni4E0B.1340uni306E.887(a)uni3068.881(b)uni306E.887uni4F8B.4011uni6587.3592uni3067.880uni306F.888uni3001.634uni540C.3209uni3058.865uni3088.913uni3046.847uni306A.883uni72B6.2525uni6CC1.1711uni304C.853uni63CF.3507uni5199.2296uni3055.862uni308C.917uni3066.879uni3044.845uni308B.916uni3053.860uni3068.881uni304C.853uni308F.920uni304B.852uni308A.915uni307E.903uni3059.866uni304C.853uni3001.634
uni666E.3539uni901A.3048uni306E.887uni65E5.3284uni672C.3722uni8A9E.1952uni6BCD.3644uni8A9E.1952uni8A71.4073uni8005.2304uni3067.880uni3057.864uni305F.872uni3089.914uni3001.634(a)uni306E.887uni307B.900uni3046.847uni304C.853uni81EA.2263uni7136.2741uni3067.880uni3042.843uni308B.916uni306E.887uni306B.884uni5BFE.2864uni3057.864(b)uni306E.887uni307B.900uni3046.847uni306F.888uni65E5.3284uni672C.3722
uni8A9E.1952uni306E.887uni6587.3592uni3068.881uni3057.864uni3066.879uni306F.888uni4E0D.3526uni81EA.2263uni7136.2741uni3067.880uni3042.843uni308B.916uni3001.634uni3068.881uni611F.1522uni3058.865uni3089.914uni308C.917uni308B.916uni306F.888uni305A.867uni3067.880uni3059.866uni3002.635
? uni307E.903uni305F.872uni3001.634(c)uni306E.887uni4F8B.4011uni6587.3592uni306E.887uni5834.2516uni5408.2041uni306F.888uni3042.843uni307E.903uni308A.915uni73FE.1905uni5B9F.2286uni3067.880uni306F.888uni8D77.1609uni304D.854uni306A.883uni3044.845uni3088.913uni3046.847uni306A.883uni72B6.2525uni6CC1.1711uni304C.853uni63CF.3507uni304B.852uni308C.917uni3066.879uni3044.845uni307E.903uni3059.866uni304C.853uni3001.634
uni305D.870uni306E.887uni4E00.1200uni65B9.3661uni3067.880uni65E5.3284uni672C.3722uni8A9E.1952uni306E.887uni6587.3592uni3068.881uni3057.864uni3066.879uni306F.888uni3054.861uni304F.856uni81EA.2263uni7136.2741uni306A.883uni3082.907uni306E.887uni3067.880uni3042.843uni308B.916uni3068.881uni8A00.1908uni3048.849uni307E.903uni3059.866uni3002.635
(a) uni305D.870uni306E.887uni7537.2953uni306E.887uni5B50.2208uni306F.888uni96A3.4001uni753A.3018uni304B.852uni3089.914uni3084.909uni3063.876uni3066.879uni304D.854uni305F.872uni8EE2.3128uni6821.1997uni751F.2652uni306E.887uni5973.2433uni306E.887uni5B50.2208uni306B.884uni604B.4032uni3092.923uni3057.864uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
(b) uni305D.870uni306E.887uni7537.2953uni306E.887uni5B50.2208uni306F.888uni96A3.4001uni753A.3018uni304B.852uni3089.914uni3084.909uni3063.876uni3066.879uni304D.854uni305F.872uni8EE2.3128uni6821.1997uni751F.2652uni306E.887uni5973.2433uni306E.887uni5B50.2208uni3092.923uni604B.4032uni3092.923uni3057.864uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
(c) uni7D2B.2229uni8272.2541uni306E.887uni6050.1706uni7ADC.3965uni304C.853uni8857.1431uni306E.887uni4E2D.2980uni3092.923uni3046.847uni308C.917uni3057.864uni305D.870uni3046.847uni306B.884uni99C6.1766uni3051.858uni56DE.1395uni3063.876uni3066.879uni3044.845uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
3. uni307E.903uni305F.872uni3001.634uni672C.3722uni8ABF.3024uni67FB.2090uni306F.888uni3042.843uni306A.883uni305F.872uni304C.853uni5B66.1462uni6821.1997uni6587.3592uni6CD5.3663uni3092.923uni3069.882uni308C.917uni3060.873uni3051.858uni6B63.2649uni3057.864uni304F.856uni899A.1454uni3048.849uni3066.879uni3044.845uni308B.916uni304B.852uni3092.923uni30C6.962uni30B9.949uni30C8.964uni3059.866uni308B.916uni3082.907uni306E.887uni3067.880uni3082.907
uni3042.843uni308A.915uni307E.903uni305B.868uni3093.924uni3002.635
203
uni4F8B.4011uni3048.849uni3070.889uni5B66.1462uni6821.1997uni306E.887uni56FD.2051uni8A9E.1952uni306E.887uni6388.2340uni696D.1728uni3067.880uni306F.888uni3001.634uni300C.686uni3089.914uni629C.3400uni304D.854uni8A00.1908uni8449.3903uni300D.687uni306F.888uni6B63.2649uni3057.864uni304F.856uni306A.883uni3044.845uni65E5.3284uni672C.3722uni8A9E.1952uni3067.880uni3042.843uni3063.876uni3066.879uni3001.634uni300C.686uni98DF.2543uni3079.898uni308C.917uni306A.883
uni3044.845uni300D.687uni3068.881uni3044.845uni3046.847uni306E.887uni306F.888uni9593.1554uni9055.1191uni3044.845uni3067.880uni300C.686uni98DF.2543uni3079.898uni3089.914uni308C.917uni306A.883uni3044.845uni300D.687uni3068.881uni8A00.1908uni3046.847uni3079.898uni304D.854uni3067.880uni3042.843uni308B.916uni3001.634uni3068.881uni7FD2.2358uni308F.920uni308C.917uni305F.872uni304B.852uni3082.907uni3057.864uni308C.917uni307E.903uni305B.868
uni3093.924uni3002.635uni3057.864uni304B.852uni3057.864uni3001.634uni672C.3722uni8ABF.3024uni67FB.2090uni3067.880uni5BFE.2864uni8C61.2501uni3068.881uni3057.864uni3066.879uni3044.845uni308B.916uni306E.887uni306F.888uni65E5.3284uni672C.3722uni8A9E.1952uni3092.923uni6BCD.3644uni8A9E.1952uni3068.881uni3057.864uni3066.879uni8A71.4073uni3059.866uni3042.843uni306A.883uni305F.872uni306E.887uni8A00.1908uni8449.3903uni306E.887uni76F4.3034uni611F.1522
uni3067.880uni3042.843uni308A.915uni3001.634uni6BCD.3644uni8A9E.1952uni8A71.4073uni8005.2304uni304C.853uni81EA.2263uni7136.2741uni3067.880uni3042.843uni308B.916uni3068.881uni5224.3409uni65AD.2949uni3059.866uni308B.916uni304B.852uni3069.882uni3046.847uni304B.852uni3001.634uni307E.903uni305F.872uni306F.888uni3053.860uni3046.847uni3044.845uni3063.876uni305F.872uni6587.3592uni3092.923uni66F8.2427uni3044.845uni305F.872uni308A.915uni767A.3395uni8A71.4073
uni3059.866uni308B.916uni3053.860uni3068.881uni304C.853uni3042.843uni308B.916uni304B.852uni3001.634uni3068.881uni3044.845uni3046.847uni3053.860uni3068.881uni3067.880uni3059.866uni3002.635uni3067.880uni3059.866uni304B.852uni3089.914uni81EA.2263uni5206.3580uni306E.887uni7B54.3188uni3048.849uni306B.884uni81EA.2263uni4FE1.2548uni3092.923uni6301.2251uni3063.876uni3066.879uni56DE.1395uni7B54.3188uni3057.864uni3066.879uni304F.856uni3060.873
uni3055.862uni3044.845uni3002.635
uni307E.903uni305F.872uni8ABF.3024uni67FB.2090uni306E.887uni4E2D.2980uni3067.880uni4F7F.2198uni308F.920uni308C.917uni3066.879uni3044.845uni308B.916uni6587.3592uni306B.884uni306F.888uni975E.3463uni5E38.2519uni306B.884uni9577.3029uni3044.845uni3082.907uni306E.887uni3082.907uni3042.843uni308A.915uni307E.903uni3059.866uni304C.853uni3001.634uni9577.3029uni3044.845uni3068.881uni3044.845uni3046.847uni7406.3943uni7531.3869uni3060.873uni3051.858
uni3067.880uni5224.3409uni65AD.2949uni305B.868uni305A.867uni306B.884uni3001.634uni3057.864uni3063.876uni304B.852uni308A.915uni6CE8.2987uni610F.1177uni3057.864uni3066.879uni6587.3592uni3092.923uni8AAD.3233uni3093.924uni3067.880uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635uni4F8B.4011uni3048.849uni3070.889uni4E0B.1340uni306E.887(d)uni306E.887uni4F8B.4011uni6587.3592uni306F.888uni308F.920uni308A.915uni3068.881uni8907.3571
uni96D1.2166uni3067.880uni3059.866uni304C.853uni3001.634uni5B9F.2286uni969B.2125uni8AAD.3233uni3080.905uni3068.881uni3054.861uni304F.856uni81EA.2263uni7136.2741uni306A.883uni6587.3592uni3067.880uni3042.843uni308B.916uni3053.860uni3068.881uni304C.853uni308F.920uni304B.852uni308A.915uni307E.903uni3059.866uni3002.635
(d) uni8ECD.1801uni968A.2882uni306E.887uni5E79.1520uni90E8.3558uni304C.853uni653F.2643uni5E9C.3534uni306E.887uni65B9.3661uni91DD.2577uni306B.884uni53CD.3411uni5BFE.2864uni3057.864uni3001.634uni8ECD.1801uni4E8B.2244uni529B.3991uni3092.923uni6226.2707uni4E89.2794uni306B.884uni5099.3467uni3048.849uni3066.879uni5897.2815uni5F37.1703uni3055.862uni305B.868uni308B.916uni3079.898uni304D.854uni3067.880uni3042.843uni308B.916uni3068.881
uni4E3B.2323uni5F35.3009uni3057.864uni3066.879uni3044.845uni308B.916uni3068.881uni3044.845uni3046.847uni30CB.967uni30E5.993uni30FC.660uni30B9.949uni306B.884uni95A2.1556uni3057.864uni3066.879uni3001.634uni5927.2887uni7D71.3191uni9818.3990uni306F.888uni3001.634uni653F.2643uni5E9C.3534uni3068.881uni8ECD.1801uni968A.2882uni9593.1554uni3067.880uni306E.887uni5BFE.2864uni8A71.4073uni306F.888uni7740.2979uni3005.657uni3068.881
uni9032.2576uni3093.924uni3067.880uni304A.851uni308A.915uni3001.634uni8FD1.1753uni65E5.3284uni4E2D.2980uni306B.884uni5408.2041uni610F.1177uni304C.853uni5F97.3224uni3089.914uni308C.917uni308B.916uni306F.888uni305A.867uni3067.880uni3042.843uni308B.916uni3068.881uni8AAC.2694uni660E.3788uni3057.864uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
uni305D.870uni308C.917uni3067.880uni306F.888uni672C.3722uni5B9F.2286uni9A13.1895uni306B.884uni9032.2576uni3080.905uni524D.2738uni306B.884uni3001.634uni4EE5.1166uni4E0B.1340uni306B.884uni3044.845uni304F.856uni3064.877uni304B.852uni7DF4.4037uni7FD2.2358uni554F.3824uni984C.2890uni304C.853uni3042.843uni308B.916uni306E.887uni3067.880uni56DE.1395uni7B54.3188uni3057.864uni3066.879uni307F.904uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845:
(e) uni3044.845uni3064.877uni305D.870uni306E.887uni5B66.1462uni751F.2652uni306F.888uni5148.2700uni751F.2652uni304C.853uni5BBF.2387uni984C.2890uni3092.923uni51FA.2394uni3057.864uni305F.872uni3068.881uni805E.3593uni3044.845uni305F.872uni306E.887?
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
(f) uni8AB0.2925uni3082.907uni304C.853uni4F55.1343uni3092.923uni98DF.2543uni3079.898uni306A.883uni304B.852uni3063.876uni305F.872uni304B.852uni305D.870uni306E.887uni30B7.947uni30A7.931uni30D5.977uni306F.888uni30A6.930uni30A8.932uni30FC.660uni30BF.955uni30FC.660uni306B.884uni805E.3593uni3044.845uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
(g) uni3069.882uni306E.887uni65B0.2557uni3057.864uni3044.845uni30CF.971uni30A4.928uni30D6.978uni30EA.998uni30C3.959uni30C9.965uni8ECA.2306uni3092.923uni304A.851uni5BA2.1644uni3055.862uni3093.924uni304C.853uni8A66.2236uni904B.1249uni8EE2.3128uni3057.864uni305F.872uni304B.852uni3069.882uni3046.847uni304B.852uni8CA9.3426uni58F2.3354uni5E97.3123uni306E.887uni5E97.3123uni4E3B.2323uni306F.888uni77E5.2956uni308A.915uni305F.872
uni304C.853uni3063.876uni3066.879uni3044.845uni305F.872uni3002.635
(uni4E0D.3526uni81EA.2263uni7136.2741) 1 2 3 4 5 6 7 (uni81EA.2263uni7136.2741)
uni6700.2103uni5F8C.1945uni306B.884uni3001.634uni5404.1444uni6587.3592uni306E.887uni81EA.2263uni7136.2741uni3055.862uni3092.923uni5224.3409uni65AD.2949uni3059.866uni308B.916uni5834.2516uni5408.2041uni3001.634uni3042.843uni307E.903uni308A.915uni8003.2015uni3048.849uni8FBC.2064uni307E.903uni305A.867uni306B.884uni76F4.3034uni611F.1522uni3092.923uni751F.2652uni304B.852uni3057.864uni3066.879uni7B54.3188uni3048.849uni308B.916uni3088.913uni3046.847
uni306B.884uni3057.864uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635uni305D.870uni308C.917uni3067.880uni306F.888uni6E96.2410uni5099.3467uni304C.853uni3067.880uni304D.854uni6B21.2253uni7B2C.2888uni3001.634uni30AF.939uni30EA.998uni30C3.959uni30AF.939uni3057.864uni3066.879uni5B9F.2286uni9A13.1895uni3092.923uni958B.1417uni59CB.2205uni3057.864uni3066.879uni304F.856uni3060.873uni3055.862uni3044.845uni3002.635
uni6B21.2253uni306E.887uni30DA.982uni30FC.660uni30B8.948uni304B.852uni3089.914uni306F.888uni5B9F.2286uni969B.2125uni306B.884uni6570.2618uni5B57.2248uni306B.884uni30AF.939uni30EA.998uni30C3.959uni30AF.939uni3059.866uni308B.916uni3068.881uni30C7.963uni30FC.660uni30BF.955uni304C.853uni8A18.1607uni9332.4069uni3055.862uni308C.917uni3001.634uni81EA.2263uni52D5.3208uni7684.3108uni306B.884uni6B21.2253uni306E.887uni6587.3592uni304C.853uni63D0.3087uni793A.2260
uni3055.862uni308C.917uni307E.903uni3059.866uni3002.635uni6700.2103uni5F8C.1945uni306B.884Thanks!uni3068.881uni3044.845uni3046.847uni30E1.989uni30C3.959uni30BB.951uni30FC.660uni30B8.948uni304C.853uni3067.880uni3066.879uni304D.854uni305F.872uni3089.914uni305D.870uni306E.887uni6642.2252uni70B9.3130uni3067.880uni5B9F.2286uni9A13.1895uni306F.888uni7D42.2356uni4E86.3971uni3057.864uni307E.903uni3059.866uni3002.635
204
Appendix B
Items
B.1 Experiment 1
(1) Template
a. Was
What
f?ur
for
eine
a
$Noun1
$Noun1
hat
has
denn
PRT
den
the
$Noun2
$Noun2
$Verb?
$Verb
b. Was
What
f?ur
for
einen
a
$Noun1
clear
hat
has
denn
PRT
die
the
$Noun2
$Noun1
$Verb?
$Verb
c. Was
What
hat
has
denn
PRT
f?ur
for
eine
an
$Noun1
$Noun1
den
the
$Noun2
$Noun2
$Verb?
$Verb
d. Was
What
hat
has
f?ur
for
eine
an
$Noun1
$Noun1
denn
PRT
den
the
$Noun2
$Noun2
$Verb?
$Verb
e. Was
What
hat
has
denn
PRT
die
the
$Noun1
$Noun1
f?ur
for
einen
a
$Noun2
$Noun2
$Verb?
$Verb
f. Was
What
hat
has
denn
PRT
f?ur
for
einen
a
$Noun2
$Noun2
die
the
$Noun1
$Noun1
$Verb?
$Verb
(2) Lexicalizations
a. $Noun1
(1) K?afer(beetle), (2)Insekt (insect), (3)Katze(cat), (4)Lehrer (teacher),
(5) Journalist (journalist), (6) Politiker (politican), (7) K?unstler (artist),
(8) Unternehmer (entrepreneur), (9) Sammler (collector), (10) Raubtier
(wild animal), (11) Sportler (athlete), (12) Handwerker (carpenter), (13)
Arzt (doctor), (14) Professor (professor), (15) B?ar (bear), (16) H?andler
(merchant), (17) Musiker (musican), (18) Hund (dog)
b. $Noun2
(1) Beamte (clerk), (2) Affe (monkey), (3) Teppich (carpet), (4) Buch
(book), (5) Artikel (article), (6) Vertrag (contract), (7) Geb?aude (build-
ing), (8) Jacht (yacht), (9)Gem?alde(painting), (10)Forscher (researcher),
(11) Turnier (tournament), (12) Schaden (damage), (13) Haustier (pet),
(14) Vortrag (presentation), (15) Fisch (fish), (16) Produkt (product),
(17) Roman (novel), (18) Knochen (bone)
c. $Verb
(1) beisen (bite), (2) stechen (sting), (3) kratzen (scratch), (4) lesen
205
(read), (5) schreiben (write), (6) unterschreiben (sign), (7) bemalen
(paint), (8) kaufen (buy), (9) ersteigern (buy at an auction), (10) an-
greifen (attack), (11) gewinnen (win), (12) reparieren (repair), (13) un-
tersuchen (investigate), (14) halten (hold), (15) fangen (catch), (16)
verkaufen (sell), (17) vertonen (set to music), (18) vergraben (bury)
B.2 Experiment 2
(3) Template
a. Gestern
yesterday
wurde
was
einem
a.DAT
$Noun1
$Noun1
ein
a.NOM
$Noun2
$Noun2
$Verb.
$Verb
b. Gestern
yesterday
wurde
was
ein
a.NOM
$Noun2
$Noun2
einem
a.DAT
$Noun1
$Noun1
$Verb.
$Verb
c. Was
what
wurde
was
denn
PRT
einem
a.DAT
$Noun1
$Noun1
f?ur
for
ein
a.NOM
$Noun2
$Noun2
$Verb?
$Verb
d. Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
$Noun1
$Noun1
ein
a.NOM
$Noun2
$Noun2
$Verb?
$Verb
e. Was
what
wurde
was
denn
PRT
f?ur
for
ein
a.NOM
$Noun2
$Noun2
einem
a.DAT
$Noun1
$Noun1
$Verb?
$Verb
f. Was
what
wurde
was
denn
PRT
ein
a.NOM
$Noun2
$Noun2
f?ur
for
einem
a.DAT
$Noun1
$Noun1
$Verb?
$Verb
g. Was
what
wurde
was
denn
PRT
f?ur
for
einem
a.DAT
$Noun1
$Noun1
f?ur
for
eine
a.NOM
$Noun2
$Noun2
$Verb?
$Verb
h. Was
what
wurde
was
denn
PRT
f?ur
for
eine
a.NOM
$Noun2
$Noun2
f?ur
for
einem
a.DAT
$Noun1
$Noun1
$Verb?
$Verb
(4) Lexicalizations
a. $Noun1
(1) Pr?uferin (examiner.fem), (2) Dozentin (lecturer.fem), (3) Offizier
(officer), (4) Babysitter (baby sitter), (5) Kommisar (detective), (6)
Polizistin (policewoman), (7)Krimineller (criminal), (8)Freundin(friend.fem),
(9) Mitarbeiter (co-worker), (10) K?onig (king), (11) Trainer (trainer),
(12) Psychiater (psychiatrist), (13) Regisseur (director), (14) Malerin
(painter.fem), (15) Arbeitsloser (unemployed), (16) Sozialarbeiterin (so-
cial worker.fem), (17) Operndirektor (impresario), (18) Zeugin (wit-
ness), (19) Minister (minister), (20) Kaiser (emperor), (21) Lehrerin
(teacher.fem), (22) Arzt (doctor), (23) M?order (murderer), (24) Profes-
sor (professor)
b. $Noun2
(1) Kandidat (candidate), (2) Sekr?aterin (secretary.fem), (3) Soldat (sol-
dier), (4)Kleinkind (baby), (5)Verbrecher (criminal), (6)Geisel (hostage),
(7) Anwalt (attorney), (8) Baby (baby), (9) Assistent (assistant), (10)
206
Bote (messenger), (11) Spieler (player), (12) Patient (patient), (13)
Schauspielerin (actress), (14)Model (model), (15) Beraterin(advisor.fem),
(16) Obdachloser (homeless person), (17) S?angerin (singer.fem), (18)
Verd?achtiger (suspect), (19) Beamte (clerk), (20) Frau (woman), (21)
Sch?uler (student), (22) Patientin (patient.fem) (23) Pflichtverteidiger
(court-appointed lawyer), (24) Student (student)
c. $Verb
(1) zeigen (show), (2) in Aussicht stellen (hold out), (3) melden (report
to), (4) anvertrauen (confide in), (5) in die Arme treiben (chase into the
arms of), (6) ausliefern (extradite), (7) vermitteln (put in contact with),
(8) ?uberlassen (leave to), (9) zuteilen (assign), (10) schicken (send), (11)
empfehlen (recommend), (12) ?ubergeben (hand over), (13) vorschlagen
(propose to), (14) pr?asentieren, (15) zuweisen (allot), (16) in Betreuung
geben (to give in custody), (17) streitig machen (to challenge for), (18)
gegen?uberstellen (confront with), (19) entziehen (deprive), (20) anbieten
(offer), (21) beschreiben (describe to), (22) ?uberweisen (refer to), (23)
zulosen (draw with), (24) vorstellen (introduce)
B.3 Experiment 3
B.3.1 Experiment 3A
(5) a. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
$Noun1
$Noun1
$Verb1
$Verb1
ist.
is.
b. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
ein
a
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
$Verb1
$Verb1
ist.
is.
c. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
einen
a
$Noun2
pilot
der
the
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
$Verb2
$Verb2
hat.
has
d. Ich
I
frage
ask
mich,
myself
was
what
der
the
$Noun1
man
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
einen
a
$Noun2
$Verb2
$Verb2
$Verb2
hat.
has
e. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
ein
a
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
$Verb2
$Verb2
wurde.
was
207
f. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
$Noun1
$Noun1
$Verb2
$Verb2
wurde.
was
g. Ich
I
frage
ask
mich,
myself
was
what
f?ur
for
ein
a
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
den
the
$Noun2
$Noun2
$Verb2
$Verb2
hat
has
h. Ich
I
frage
ask
mich,
myself
was
what
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
$Noun1
$Noun1
den
the
$Noun2
$Noun2
$Verb2
$Verb2
hat
has
i. Ich
I
frage
ask
mich,
myself
was
what
f?ur
yesterday
ein
on.the
Mann
afternoon
gestern
for
am
a
Nachmittag
man
$Verb3
$Verb3
hat.
has
(6) Lexicalizations
a. $Noun1
(1) Mann (man), (2) Frau (woman), (3) Detektiv (detective), (4) Pen-
sionist (pensioner), (5) Handwerker (carpenter), (6) Vogel (bird), (7)
Ganove (crook), (8) Politiker (politician), (9) Schauspieler (actor), (10)
Sportler (athlete), (11) Musiker (musican), (12) Adeliger (noble man),
(13)Polizist (policeofficer), (14) Maler (painter), (15)Autofahrer (driver),
(16) Radfahrer (bicyclist), (17) Arzt (doctor), (18) Student (student),
(19) Sch?uler (student), (20) Verwandter (relative), (21) Anwalt (attor-
ney), (22) Soldat (soldier), (23) Sozialarbeiter (social worker), (24) Kell-
ner (waiter), (25) L?owe, (26) Sekr?ater (secretary), (27) Rennfahrer (race
car driver), (28) Sammler (collector), (29) Kommisar (inspector), (30)
Tiger (tiger),
b. $Noun2
(1) Pilot (pilot), (2) Mitarbeiter (co-worker), (3) Dieb (thief), (4) Nach-
bar (neighbor), (5) Architekt (architect), (6) K?afer (beetle), (7) Passan-
ten (bystander), (8) Minister (minister), (9) Regisseur (director), (10)
Journalist (journalist), (11) S?anger (singer), (12) Pudel (poodle), (13)
Demonstrant (demonstrator), (14) Prinz (prince), (15) Anhalter (hitch-
hiker), (16) Fu?g?anger (pedestrian), (17) Patient (patient), (18) Haus-
meister (janitor), (19) Lehrer (teacher), (20) Hund (dog), (21) Richter
(judge), (22) Zivilist (civilian), (23) Obdachlosen (homeless person), (24)
Gast (guest), (25) B?uffel (buffalo), (26) Buchhalter (accoutant), (27)
T?anzer (dancer), (28) K?unstler (artist), (29) Verd?achtiger (suspect), (30)
Hirsch (deer)
208
c. $Verb1
(1) erscheinen (appear), (2) kommen (come), (3) einschlafen (fall asleep),
(4) einnicken (doze off), (5) stolpern (stumble), (6) schl?upfen (hatch), (7)
auftauchen (surface), (8) ausrutschen (slip), (9) sterben (die), (10) aus-
fallen (drop out), (11) zusammenbrechen (collapse), (12) aufschrecken
(startle up), (13) st?urzen (fall), (14) verunfallen (have an accident),
(15) aufwachen (wake up), (16) umkommen (die in an accident), (17)
abst?urzen (crash), (18) scheitern (fail), (19) ankommen (arrive), (20)
vorbeikommen (drop by), (21) fallen (die in war), (22) verschwinden
(vanish), (23) eintreffen (arrive), (24) entwischen (escape), (25) her-
vorstechen (stand out), (26) verungl?ucken (have an accident), (27) (28)
ausrasten (freak out), (29) straucheln (falter), (30) entkommen (escape)
d. $Verb2
(1) sehen (see), (2)entlassen (fire), (3) erwischen (catch), (4)beschimpfen
(insult), (5) verst?andigen (contact), (6) fressen (eat), (7) ?uberfallen
(mug), (8) bestechen (bribe), (9) kritisieren (criticize), (10) ignorieren
(ignore), (11) begleiten (accompany), (12) auslachen (laugh at), (13)
zurechtweisen (reprimand), (14) portraitieren (portray), (15) mitnehmen
(bring along), (16) ?uberholen (overtake), (17) behandeln (treat), (18)
bel?astigen (molest), (19) zwicken (pinch), (20) streicheln (pet), (21)
einladen (invite), (22) retten (save), (23) betreuen (advise), (24) be-
dienen (wait on), (25) ersp?ahen (spot), (26) kritisieren (criticize), (27)
entdecken (discover), (28) unterst?utzen (support), (29) verh?oren (inter-
rogate), (30) jagen (hunt)
e. $Verb3
(1) schlafen (sleep), (2) schnarchen (snore), (3) weinen (cry), (4) lachen
(laugh), (5) husten (cough), (6) zwitschern (twitter), (7) singen (sing),
(8) tanzen (dance), (9) musizieren (play music), (10) teilnehmen (partic-
ipate), (11) spielen (play), (12) sprechen (talk), (13) trinken (drink), (14)
rauchen (smoke), (15) bremsen (hit the breaks), (16) g?ahnen (yawn),
(17) operieren (perform surgery), (18) lernen (study), (19) lesen (read),
(20) basteln (do craft), (21) l?ugen (lie), (22) angreifen (attack), (23)
helfen (help), (24) abkassieren (cash up), (25) dosen (dose), (26) tele-
fonieren (be on the phone), (27) niesen (sneeze), (28) mitbieten (partic-
ipate in an auction), (29) nachdenken (ponder), (30) fressen (eat)
B.3.2 Experiment 3B
(7) a. Was
what
f?ur
for
$Noun1
$Noun1
sind
are
denn
PRT
am
on.the
Nachmittag
afternoon
$Verb1?
$Verb1
209
b. Was
what
sind
are
denn
PRT
f?ur
for
$Noun1
$Noun1
am
on.the
Nachmittag
afternoon
$Verb1?
$Verb1
c. Was
what
f?ur
for
$Noun1
$Noun1
haben
have
denn
PRT
am
on.the
Nachmittag
afternoon
$Verb2?
$Verb2
d. Was
what
haben
have
denn
PRT
f?ur
for
$Noun1
$Noun1
am
on.the
Nachmittag
afternoon
$Verb2?
$Verb2
(8) Lexicalizations
a. $Noun1
(1) M?anner (men), (2) Frauen (women), (3) M?adchen (girls), (4) Stu-
denten (students), (5) Sch?ulerinnen (students.fem), (6) Radfahrerinnen
(cyclists.fem), (7) Ganoven (crooks), (8) Sportlerinnen (athletes.fem),
(9) Gefange (prisoners), (10) Politiker (policitian), (11) Musikerinnen
(musicians), (12) Kinder (children)
b. $Verb1
(1) ankommen (arrive), (2) abreisen (depart), (3) weggehen (leave), (4)
zur?uckkommen (return), (5) davonlaufen (run away), (6) davonfahren
(drive away), (7) fliehen (flee), (8) mitlaufen (participate in a race),
(9) entkommen (escape), (10) auftauchen (appear), (11) auftreten (per-
form), (12) weglaufen (run away)
c. $Verb2
(1) arbeiten (work), (2) singen (sing), (3) warten (wait), (4) reden (talk),
(5) telefonieren (talk on the phone), (6) plaudern (chat), (7) aufgeben
(give up), (8) mitspielen (play along), (9) kooperieren (cooperate), (10)
verhandeln (negotiate), (11) spielen (play), (12) tratschen (gossip)
B.4 Experiment 4
(9) a. $MatrixPredicate,
$MatrixPredicate
was
what
f?ur
for
ein
a
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
den
the
$Noun2
$Noun2
$Verb
$Verb
hat.
has
b. $MatrixPredicate,
$MatrixPredicate
was
what
f?ur
for
einen
a
$Noun2
$Noun2
der
the
$Noun1
$Noun1
gestern
yesterday
am
on.the
Nachmittag
afternoon
$Verb
$Verb
hat.
has
c. $MatrixPredicate,
$MatrixPredicatewhat
was
yesterday
gestern
on.the
am
afternoon
Nachmittag
for
f?ur
a
ein
$Noun1
$Noun1
the
den
$Noun2
$Noun2
$Verb
$Verb
has
hat.
210
d. $MatrixPredicate
$MatrixPredicate
was
what
den
the
$Noun2
clerk
gestern
yesterday
am
on.the
Nachmittag
afternoon
f?ur
for
ein
a
$Noun1
$Noun1
$Verb
bitten
hat.
has
e. Ich
$MatrixPredicatewhat
frage
the
mich,
$Noun1
was
yesterday
der
on.the
$Noun1
afternoon
gestern
for
am
a
Nachmittag
$Noun2
f?ur
$Verb
einen
has
$Noun2 $Verb hat.
(10) Lexicalizations
a. $MatrixPredicate
(1) ich frage mich (I ask myself), (2) ich bin nicht sicher (I am not
sure), (3) ich wei? nicht (I do not know), (4) ich m?ochte wissen (I
would like to know), (5) es ist unklar (it is unclear), (6) ich habe mich
erkundigt (I inquired), (7) ich habe nachgefragt (I asked), (8) ich wollte
herausfinden (I wanted to find out), (9) ich frage mich (I ask myself),
(10) ich habe nachgefragt (I asked), (11) man kann nicht sagen (one
can?t tell), (12) ich wei? nicht (I do not know), (13) ich m?ochte wissen
(I would like to know), (14) es ist unklar (it is unclear), (15) ich habe
mich erkundigt (I inquired)
b. $Noun1
(1) K?afer (beetle), (2) Affe (monkey), (3) Kater (male cat), (4) Lehrer
(teacher), (5) Sportler (athlete), (6) Teamleiter (team leader), (7) Vis-
agist (make-up artist), (8) Tiger (tiger), (9) Arzt (doctor), (10) Polizist
(police officer), (11) Vogel (bird), (12) Sammler (collector), (13) Ar-
chitekt (architect), (14) Reporter (reporter), (15) Student (student)
c. $Noun2
(1) Beamte (clerk), (2) Fisch (fish), (3) Hund (dog), (4) Sch?uler (stu-
dent), (5)Journalist (journalist), (6)Mitarbeiter (co-worker), (7) Schaus-
pieler (actor), (8) Forscher (researcher), (9) Patient (patient), (10) Ver-
brecher (criminal), (11) Wanderer (hiker), (12) K?unstler (artist), (13)
Handwerker (carpenter), (14) Star (star), (15) Professor (professor)
d. $Verb1
(1) bei?en (bite), (2) fressen (eat), (3) kratzen (scratch), (4) pr?ufen
(test), (5) beschimpfen (berate), (6) entlassen (fire), (7) schminken
(put make-up on), (8) attackieren (attack), (9) untersuchen (examine),
(10) verfolgen (chase), (11) verletzen (hurt), (12) untersch?atzen (un-
derestimate) (13) verst?andigen (notify), (14) beleidigen (insult), (15)
bel?astigen (molest)
211
B.5 Experiment 5
(11) a. DET
The
$Noun2
$Noun2
zu
to
$Verb
$Verb
hat
has
DET
the
$Noun1
$Noun1
$MatrixPredicateA.
$MatrixPredicateA
b. Welche/n
Which
$Noun2
$Noun2
hat
has
denn
PRT
zu
to
$Verb
$Verb
DET
the
$Noun1
$Noun1
$MatrixPredicateA?
$MatrixPredicateA
c. DET
The
$Noun1
$Noun1
hat
has
DET
the
$Noun2
$Noun2
zu
to
$Verb
$Verb
$MatrixPredicateB.
$MatrixPredicateB
d. Welche/n
Which
$Noun2
$Noun2
hat
has
denn
PRT
DET
the
$Noun1
$Noun1
zu
to
$Verb
$Verb
$MatrixPredicateB?
$MatrixPredicateB
e. Es
it
hat
has
DET
the
$Noun1
$Noun1
$MatrixPredicateA
$MatrixPredicateA
DET
the
$Noun2
$Noun2
zu
to
$Verb.
$Verb
f. Welche/n
Which
$Noun2
$Noun2
hat
has
(es)
(it)
denn
PRT
DET
the
$Noun1
$Noun1
$MatrixPredicateA
$MatrixPredicateA
zu
to
$Verb?
$Verb
g. DET
The
$Noun1
$Noun1
hat
has
$MatrixPredicateB
$MatrixPredicateB
DET
the
$Noun2
$Noun2
zu
to
$Verb.
$Verb
h. Welche
Which
$Noun2
$Noun2
hat
has
denn
PRT
DET
the
$Noun1
$Noun1
$MatrixPredicateB
$MatrixPredicateB
zu
to
$Verb?
$Verb
(12) Lexicalizations
a. $Noun1
(1) Studentin (student.fem), (2) Sozialarbeiterin (social worker.fem),
(3) Praktikant (intern), (4) Regisseur (director), (5) Kind (child), (6)
Mann (man), (7) Priester (priest), (8) Architektin (architect.fem), (9)
Journalist (journalist), (10) Pianist (pianist), (11) Immobilienh?andler
(real estate agent), (12) Polizist (police officer)
b. $Noun2
(1) Diplomarbeit (master thesis), (2) Asylwerber, (3) Patient (patient),
(4) Schauspieler (actor), (5) Esel (donkey), (6) Badezimmer (bath-
room), (7) Bischof (bishop), (8) Projekt (project), (9) Artikel (arti-
cle), (10) Sonate (sonata), (11) Grundst?uck (property), (12) Geb?aude
(building)
212
c. $Verb
(1) schreiben (write), (2) helfen (help), (3) behandeln (treat), (4) un-
terst?utzen (support), (5) streicheln (pet), (6) putzen (clean), (7) kri-
tisieren (criticise), (8) abbrechen (abort), (9) schreiben (write), (10)
r?uckw?arts zuspielen (play backwards), (11)verkaufen (sell), (12)r?aumen
(clear)
d. $MatrixPredicateA
(1) vorhaben (plan), (2) erw?agen (consider), (3) verabs?aumen (fail),
(4) verlautbaren (announce), (5) anfangen (begin), (6) versprechen
(promise), (7) wagen (dare), (8) erzwingen (force), (9) im Sinn haben
(have in mind), (10) probieren (try), (11) in Erw?agung ziehen (take
into consideration), (12) veranlassen (order)
e. $MatrixPredicateB
(1)langweilen (bore), (2)Problem bereiten(cause problems), (3) ver?arg-
ern (annoy), (4) belasten (burden), (5) erfreuen (delight), (6) entz?uckt
(delighted), (7) das Amt kosten (cost his office), (8) emp?ort (appall), (9)
Zeit kosten (cost time), (10) anstrengen (exhaust), (11) M?uhe machen
(make an effort), (12) Schwierigkeiten machen (cause problems)
B.6 Experiment 6
(13) a. Welche/n
which
$Noun1
$Noun1
hat
has
denn
PRT
DET
the
$Noun2
$Noun2
zu
to
$Verb
$Verb
$MatrixPredicateA?
$MatrixPredicateA
b. Welche/n
which
$Noun1
$Noun1
$MatrixPredicateA
$MatrixPredicateA
denn
PRT
DET
the
$Noun2
$Noun2
zu
to
$Verb?
$Verb
c. Welche/n
which
$Noun1
$Noun1
hat
has
denn
PRT
DET
the
$Noun2
$Noun2
zu
to
$Verb
$Verb
$MatrixPredicateB?
$MatrixPredicateB
d. Welche/n
which
$Noun1
$Noun1
$MatrixPredicateB
$MatrixPredicateB
denn
PRT
DET
the
$Noun2
$Noun2
zu
to
$Verb?
$Verb
(14) Lexicalizations
a. $Noun1
213
(1)Sonate(sonata), (2)Arbeit (paper), (3) Patient (patient), (4) Schaus-
pieler (actor), (5) Esel (donkey), (6) Zimmer (room), (7) Projekt (project),
(8)Artikel (article), (9)Lied(song), (10) Bischof (bishop), (11)Rotwein
(red wine), (12) Patienten (patient)
b. $Noun2
(1) Pianist (pianist), (2) Studentin (student.fem), (3) Praktikant (in-
tern), (4) Regisseur (director), (5) Kind (child), (6) Mann (man), (7)
Architekt (architect), (8)Journalistin(journalist.fem), (9)Tante (aunt),
(10) Priester (priest), (11) Weinkennerin (wine enthusiat.fem), (12)
Praktikantin (intern.fem)
c. $Verb
(1) spielen (play), (2) schreiben (write), (3) behandeln (treat), (4) un-
terst?utzen (support), (5) streicheln (pet), (6) putzen (clean), (7) been-
den (end), (8) schreiben (write), (9) belustigen (amuse), (10) kritisieren
(criticise), (11) verkosten (taste), (12) helfen (help)
d. $MatrixPredicateA
(1) forden (challenge), (2) langweilen (bore), (3) ver?argern (annoy), (4)
belasten (burden), (5) erfreuen (delight), (6) entz?uckt (charme), (7)
emp?oren (appall), (8) interessieren (interest), (9) belustigen (amuse),
(10) erleichtern (unburden), (11) begl?ucken (satisfy), (12) bereichern
(enrich)
e. $MatrixPredicateA
(1) forden (demand), (2) planen (plan), (3) verabs?aumen (fail), (4) ver-
lautbaren (announce), (5) beginnen (begin), (6) versprechen (promise),
(7) erzwingen (force), (8) erw?agen (consider), (9) hoffen (hope), (10)
wagen (dare), (11) entscheiden (decide), (12) probieren (try)
B.7 Experiment 7
(15) a. DET
the
$Noun1
$Noun1
zu
to
$Verb
$Verb
$MatrixPredicateA
$MatrixPredicateA
DET
the
$Noun2
$Noun2
PRT.
PRT
b. Welche/n
which
$Noun1
$Noun1
$MatrixPredicateA
$MatrixPredicateA
denn
PRT
zu
to
$Verb
$Verb
die
the
$Noun2
$Noun2
PRT?
PRT
c. DET
the
$Noun1
$Noun1
zu
to
$Verb
$Verb
$MatrixPredicateB
$MatrixPredicateB
DET
the
Studentin.
$Noun2
214
d. Welche/n
which
$Noun1
$Noun1
$MatrixPredicateB
$MatrixPredicateB
denn
PRT
zu
to
$Verb
$Verb
DET
the
$Noun2?
$Noun2
a. $Noun1
(1) Lied (song), (2) Arbeit (paper), (3) Zaun (fence), (4) Garten (garden),
(5) Dachstuhl (truss), (6) Christbaum (Christmas tree), (7) Brand (fire),
(8) Feld (field), (9) Schl?ager (racket), (10) Kasten (cupboard), (11) Kran
(crane), (12) K?uhlschrank (fridge)
b. $Noun2
(1) Musikerin (musican.fem), (2) Studentin (student.fem), (3) Heimwerk-
erin (home improver.fem), (4) Pensionistin (pensioner.fem), (5) Dachdecker
(roofer), (6) Onkel (uncle), (7) Feuerwehrmann (fire fighter), (8) Landwirten
(farmer.fem), (9) Tennisspielerin (tennis player.fem), (10) Mieter (tenant),
(11) Kranfhrer (crane operator), (12) Choleriker (choleric)
c. $Verb
(1) spielen (play), (2) schreiben (write), (3) streichen (paint), (4) pflegen
(take care of), (5) bauen (build), (6) schm?ucken (decorate), (7) l?oschen
(extinguish), (8) pfl?ugen (write), (9) bespannen (string), (10) zertr?ummern
(demolish), (11) lenken (manoeuvre), (12) einschlagen (smash)
d. $MatrixPredicateA
(1) anstrengen (exhaust), (2)aufregen (annoy), (3) schwerfallen (is difficult),
(4) anregen (encourage), (5) auslaugen (wear out), (6) ausbauen (extend),
(7) mitnehmen (exhaust), (8) aufheitern (cheer up), (9) ablenken (distract),
(10) aufw?uhlen (stir up), (11) zusetzen (badger), (12) nachh?angen (dwell on)
e. $MatrixPredicateB
(1) belasten (burden), (2) ver?argen (annoy), (3) missfallen (displease), (4)
motivieren (motivate), (5) erm?uden (exhaust), (6) liegen (suit), (7) er-
sch?opfen (wear down), (8) erfreuen (please), (9) st?oren (bother), (10) en-
tkr?aften (debilitate), (11) verunsichern (unsettle), (12) schaden (harm)
B.8 Experiment 8
(16) Template
a. John wondered whether a book about Obama had created a scandal
last year.
b. John wondered about which politician a book had created a scandal
last year.
215
c. John wondered which politician a book about had created a scandal
last year.
d. John wondered whether Mary had read a book about Obama last year.
e. John wondered which politician Mary had read a book about last year.
f. John wondered about which politician Mary had read a book last year.
(17) Lexicalizations (for all conditions according to template)
1 Anne speculated whether a magazine about {gardening, which topic}
had helped the readers yesterday.
2 Tom wondered whether a documentary about {poverty, which issue}
had motivated the viewers last week.
3 Monica wondered whether a presentation about {psychology, which
subject} had informed the audience on Tuesday.
4 Bob wondered whether a musical about {cats, which animal} had got-
ten good reviews last year.
5 Kate wondered whether a novel about {Catholics, which religion} had
caused a scandal last month.
6 Nick wondered whether a song about {love, which emotion} had in-
spired the listeners this morning.
7 Phil speculated whether adocumentary about{healthcare, which topic}
had swayed the voters last year.
8 Zach wondered whether a TV show about {sharks, which animal} had
excited the viewers yesterday.
9 Alex wondered whether a class about{global warming, which problem}
had inspired the students last semester.
10 Danielle inquired whether a press release about {Palestine, which coun-
try} had cleared up the controversy last week.
11 Hannah wondered whether an essay about {education, which question}
had impressed the teacher today.
12 Bob speculated whether a protest about {democracy, which human
right} had influenced the politicians on Monday.
13 Jack wondered whether a story about {tigers, which animal} had ex-
cited the readers yesterday.
14 John wondered whether a song about {love, which feeling} had caused
a revolution last night.
15 Craig wondered whether a race for {cancer, which disease} had been a
success last year.
16 Mark inquired whether a story about {Clifford, which dog} had caused
an uproar last week.
17 Brandon wondered whether a radio program about {sex, which theme}
had offended Conservatives last night.
18 Julia wondered whether a dissertation about {the Cold War, which
war} had brought new insight last winter.
216
B.9 Experiment 9
(18) Template
a. A book about Obama caused a scandal.
b. Which politician did a book about cause a scandal?
c. John believed a book about Obama to have caused a scandal.
d. Which politician did John believe a book about to have caused a scan-
dal?
(19) Lexicalizations (for all conditions according to template)
1 Chris believed amovie about{capitalism, which subject}tohave caused
controversy.
2 Michael believed a TV show about {doctors, which topic} to have en-
tertained the viewers.
3 Nick believed a class about {American history, which field} to have
challenged the students.
4 Ryan believed a book about {Christmas, which holiday} to have cap-
tivated the children.
5 Kim believed a poem about {Paris, which city} to have delighted the
teachers.
6 Steven believed a play about {WWII, which war} to have shocked the
audience.
7 Kurt believed a magazine about {movie stars, which celebrities} to
have bored the readers.
8 Kurt believed a seminar about {crime, which topic} to have created a
strong response.
9 Janet believed a report about {Iran, which country} to have fooled the
investigators.
10 Celine believed a song about {love, which emotion} to have excited the
listeners.
11 Paul believed an article about {drugs, which issue} to have angered
people.
12 Tim believed a novel about {Mormons, which religion} to have caused
a scandal.
B.10 Experiment 10
(20) Template
a. John wondered which man a book about caused a scandal last year.
b. John wondered whether a book about Obama caused a scandal last
year.
c. John wondered which man a book about was released last year.
d. John wondered whether a book about Obama was released last year.
217
(21) Lexicalizations (for all conditions according to template)
1 Steve wondered whether a TV show about {Sports, which topic} had
{excited people, been aired} last night.
2 Mary speculated whether a website about {Obama, which politician}
had {spread rumors, been launched} last week.
3 Lucy inquired whether a movie about{lions, which animal}had{attracted
audiences, been released} last month.
4 Mike wondered whether a book about {George Bush, which man} had
{caused a scandal, been written} last year.
5 Anne speculated whether an article about {Tom Cruise, which actor}
had {angered people, been published} on Monday.
6 Joe inquired whether a documentary about {Sudan, which country}
had {sparked a revolution, been made} last week.
7 Sue wondered whether a song about {Marilyn Monroe, which woman}
had {interested people, been performed} today.
8 Robert speculated whether a story about {Peter Pan, which character}
had {gained notoriety, been read} last week.
9 Jane inquired whether a magazine about {health, which subject} had
{intrigued readers, been mailed} on Tuesday.
10 John wondered whether a lecture about {evolution, which theory} had
{fascinated people, been given} last semester.
11 Kate speculated whether a class about {genocide, which issue} had
{caused a controversy, been offered} last week.
12 Keith wondered whether a presentation about {World War II, which
topic} had {caused excitement, been delivered} this morning.
B.11 Experiment 11
(22) Template
a. A different chapter of the textbook seems to have been assigned to
John.
b. A different chapter of the textbook seems to have been assigned to
every student.
c. That?s the book that a different chapter of seems to have been assigned
to John.
d. That?s the book that a different chapter of seems to have been assigned
to every student.
e. That?s the book that John seems to have assigned a different chapter
of to every student.
(23) Lexicalizations (for all conditions according to template)
1 A different chapter of the textbook seems to have been assigned to
{John, every student}.
218
2 A different section of the manuscript seems to have been handed out
to {every reviewer, Mary}.
3 A different clip of the movie seems to have been shown to {Bill, every
member of the Academy}.
4 A different episode of the TV show seems to have been devoted to
{Einstein, every Nobel Price laureate}.
5 A different excerpt of the radio program seems to have been presented
to {Linda, every critic}.
6 A different part in the play seems to have been written for{Chris, every
actor}.
7 A different novel about {Baltimore, the city} seems to have been sold
to {Anna, every publisher}.
8 A different lecture about global warming seems to have been given to
{Michael, every CEO}.
9 A different passage of the story seems to have been read to {Carol,
every child}.
10 A different ad for a medication seems to have been emailed to {Jason,
every employee}.
11 A different report about the city budget seems to have been mailed to
{Susan, every council member}.
12 A different parking spot in {lot 1, the lot} seems to have been allocated
to {Bob, every resident}.
13 A different line of the poem seems to have been dedicated to {Barbara,
every president}.
14 A different coupon for{a free soda, the item}seems to have been offered
to {Tim, every customer}.
15 A different travel guide about {Austria, the country} seems to have
been recommended to {Jennifer, every tourist}.
B.12 Experiment 12
(24) Template
a. sono
the
$Noun1-wa
$Noun1-Top
$NP2-ga
$NP2-Nom
$NP3-o
$NP3-Acc
$Verb1
$Verb1
koto-ga
fact-Nom
$Predicate1
$Predicate1
to
Comp
$Verb2
$Verb2
b. $NP3-o
$NP3-Acc
sono
the
$Noun1-wa
$Noun1-Top
$NP2-ga
$NP2-Nom
$Verb1
$Verb1
koto-ga
fact-Nom
$Predicate1
$Predicate1
to
Comp
$Verb2
$Verb2
c. sono
the
$Noun1-wa
$Noun1-Top
$NP2-ga
$NP2-Nom
$NP3-o
$NP3-Acc
$Verb1
$Verb1
koto-o
fact-Acc
$Predicate2
$Predicate2
to
Comp
$Verb2
$Verb2
219
d. $NP3-o
$NP3-Acc
sono
the
$Noun1-wa
$Noun1-Top
$NP2-ga
$NP2-Nom
$Verb1
$Verb1
koto-o
fact-Acc
$Predicate2
$Predicate2
to
Comp
$Verb2
$Verb2
e. sono
the
$Noun1-ga
$Noun1-Nom
$NP2-ga
$NP2-Nom
$Verb1
$Verb1
koto-ga
fact-Nom
$Predicate1
$Predicate1
to
Comp
$Verb2
$Verb2
no-wa
NL-Top
$NP3
$NP3
da
cop
f. sono
the
$Noun1-ga
$Noun1-Nom
$NP2-ga
$NP2-Nom
$Verb1
$Verb1
koto-o
fact-Acc
$Predicate1
$Predicate1
to
Comp
$Verb2
$Verb2
no-wa
NL-Top
$NP3
$NP3
da
cop
(25) Lexicalizations
a. $Noun1
(1) kooti (coach), (2) kantoku (director), (3) saibankan (judge), (4)
sakka (writer), (5) butyoo (manager), (6) syouzyo (girl), (7) yakuin (ex-
ecutive), (8) kankyaku (audience), (9) hahaoya (mother), (10) sakkyo-
kuka (composer), (11) kyouzyu (professor), (12) kangohutyou (chief
nurse), (13) syatyou (president), (14) zyaanarisuto (journalist), (15)
kityou (captain), (16) kannusi (priest), (17) sigikaigiin (city council
member), (18) resutoran-no tentyou (restaurant owner)
b. $NP2
(1) mumei-no rikuzyou sensyu (nameless track-and-field athlete), (2)
oomonohyouronka (bigwig critic), (3) torakkuno untensyu (truck driver),
(4)nekkyoutekina stookaa(fanatical stalker), (5)yuusyuuna syain (com-
petent employee), (6) iziwaruna ane (mean older sister), (7) hiroti-no
buka (one of his men), (8) hawaizin-no myuuzisyan (Hawaiian mu-
sician), (9) ukkarimono-no musuko (forgetful son), (10) ninki kasyu
(popular singer), (11) zyosyu (assistant), (12) isya (doctor), (13) izi-
waruna senmu (mean senior director), (14) gakusei borantia (student
volunteer), (15) sutyuwaadesu (cabin crew), (16) miko (shrine maiden),
(17) hisyo (secretary), (18) syehu (chef)
c. $NP3
(1) 100m-no kiroku (record in 100 meters), (2) syuen haiyuu (main
actor), (3) humikiri-no syadanki (railway crossing gate), (4) kyouhaku-
no tegami (threatening letter), (5) kaigaide-no ninmu (duty abroad),
(6) kuma-no nuigurumi (teddy bear), (7) keieizyou-no himitu (secret
management information), (8) nihonsei-no ukurere (Japanese ukulele),
(9) uraguti-no doa (back door), (10) hazimete-no zenkoku tuaa (first
national tour), (11) zikken deeta (experimental data), (12) nyuugan
220
kensin (breast cancer examination), (13) sinnyuu syain (new employee),
(14) ahurika-de-no iryou enzyo (medical aid in Africa), (15) gaikokuzin-
no zyoukyaku (foreign passenger), (16) keidai-no souzi (cleaning of the
precinct), (17) yuunouna sityou (competent mayor), (18) tyuubou-no
nezumi kuzyo (extermination of the rats in the kitchen)
d. $Verb1
gwig (1) nurikaeta (renewed), (2) kokuhyousita (harshly criticized),
(3) hakai-sita (destroyed), (4) kaita (wrote), (5) kanryousita (com-
pleted), (6) suteta (dumped), (7) nigitteiru (hold), (8) hiita (played),
(9) akeppana-sinni sita (left open), (10) tyuudansita (suspended), (11)
kaizansita (falsified), (12) orosokani sita (did carelessly), (13) haku-
gaisita (persecuted), (14) sasaeteita (was supporting), (15) naigasiro-ni
sita (ignored), (16) namaketa (neglected), (17) kokuhatusita (accused),
(18) okotatta (neglected)
e. $Predicate1
(1) marukomi-o sawagaseta (caused sensation among the media), (2)
kougyou seiseki-no husin-ni tunagatta (led to sluggish box office re-
sults), (3) daiziko-o hikiokosita (caused a big accident), (4) rensai-no
utikiri-o unagasita (accelerate the novel series to the end), (5) kaigai
zigyou-o hatten saseta (expanded their overseas business), (6) kenka-no
genin da (is the cause of the fight), (7) kaisya-no sonzoku-o obiyakasita
(menaced the continuation of the company), (8) konsaato-no medama-
ni natta (became the main event of the concert), (9) akisu higai-o
maneita (caused the loss by theft), (10) fan-o situbou saseta (disap-
pointed the fans), (11) kenkyuusitu-no hyouban-o sageta (degraded
the reputation of the research lab), (12) kanzya-no byouzyou-o akka
saseta (made the patient?s condition worse), (13) syokuba-no huniki-o
gikotinaku saseta (made the atmosphere in the office awkward), (14)
kokuren hakenin-o kansin saseta (impressed the UN delegate), (15)
koukuugaisya-no ninki-o teimei saseta (lowered the popularity of the
airline), (16) hatumoode kyaku-o okorasete simatta (ended up upset-
ting the new year visitors), (17) simin-o odorokaseta (surprised the
citizens), (18) mise-no hyouban-o dainasi-ni sita (ruined the reputation
of the store)
f. $Predicate2
(1) zimansita (bragged about), (2) koukaisiteiru (is regretting), (3)
kakusiteita (was hiding), (4) zihakusita (confessed), (5) yorokondeita
(was happy about), (6) naysyo-ni siteita (kept as a secret), (7) bakuro-
sita (revealed), (8) hokotteita (was proud about), (9) wasureteiru (has
forgotten), (10) wasurerarezuni iru(still cannot forget), (11)impeisiteita
(has concealed), (12) mitometeiru (admit), (13) hiteisita (denied), (14)
kokuhakusita (confessed), (15) kuyandeiru (is regretting), (16) mou-
221
seisiteiru (repented), (17) kouhyousiteiru (has made public), (18) koku-
hakusita (confessed)
g. $Verb2
(1) kangaeta (thought), (2) suisokusita (guessed), (3) syutyousita (ar-
gued), (4) hanasita (talked), (5) komentosita (commented), (6) ut-
taeta (claimed), (7) sinzikondeita (believed), (8) katatta (told), (9)
kizuita (noticed), (10) suisokusita (guessed), (11) omotta (thought),
(12) syougensita (testified), (13) hanasiteita (was talking), (14)houdou-
sita (reported), (15) hanasita (talked), (16) omoikondeita (falsely be-
lieved), (17) setumeisita (explained), (18) kanasinda (grieved)
B.13 Experiment 13
Left Branch Extraction
(26) Template
a. ?Ciji
whose
su
aux
$ADJUNCT
$ADJUNCT
$Noun1
$Noun1
$Verb
$Verb
$Noun2
$Noun2
b. ?Ciji
whose
$Noun1
$Noun1
su
aux
$ADJUNCT
$ADJUNCT
$Verb
$Verb
$Noun2
$Noun2
c. ?Cije
whose
su
aux
$ADJUNCT
$ADJUNCT
$Noun1
$Noun1
$Verb
$Verb
$Noun2
$Noun2
d. ?Cije
whose
$Noun2
$Noun2
su
aux
$ADJUNCT
$ADJUNCT
$Noun1
$Noun1
$Verb
$Verb
(27) Lexicalizations
a. $Noun1
(1) prijatelji (friends), (2) roditelji (parents), (3) kolege (colleagues),
(4) poznanici (acquaintances), (5) ?cuvari (guards), (6) drugarice (fe-
male friends), (7) profesori (professors), (8) u?cenici (pupils), (9) stu-
denti (students), (10) lekari (doctors), (11) rodaci (cousins), (12) gazde
(landlords)
b. $Noun2
(1) kom?sije (neighbours), (2) drugari (friends), (3) plesa?ci (dancers),
(4) partneri (partners), (5) sportisti (athletes), (6) voza?ci (drivers), (7)
muzi?cari (musicians), (8) nastavnici (teachers), (9) u?citelji (teachers),
(10) instruktori (instructors), (11) de?caci (boys), (12) devoj?cice (girls)
c. $ADJUNCT
(1) pro?sle godine (last year), (2) pro?slog meseca (last month), (3) pro?sle
nedelje (last week), (4) pro?slog oktobra (last October), (5) pro?slog
222
ponedeljka (last Monday), (6) pro?slog utorka (last Tuesday), (7) pro?sle
srede (last Wednesday), (8) pro?slog?cetvrtka (last Thursday), (9)pro?slog
petka (last Friday), (10) pro?sle subote (last Saturday), (11) pro?sle
nedelje (last Sunday), (12) pro?slog leta (last summer)
d. $Verb
(1) upoznati (meet), (2) bodriti (support), (3) gledati (watch/observe),
(4)prevariti (cheat), (5) udariti(hit), (6)poljubiti (kiss), (7)podu?cavati
(teach), (8) za?cikavati (tease), (9) nacrtati (draw), (10) slagati (lie),
(11) zagrliti (hug), (12) prepoznati (recognize)
PP-Extraction
(28) Template
a. O
about
kome/?cemu
whom/what
je
aux
$ADJUNCT
$ADJUNCT
taj/onaj
this/that
$Noun1
$Noun1
$Verb
$Verb
$Noun2?
$Noun2
b. taj/onaj
this/that
$Noun1
$Noun1
je
aux
$ADJUNCT
$ADJUNCT
$Verb
$Verb
$Noun2
$Noun2
c. O
about
kome/?cemu
whom/what
je
aux
$ADJUNCT
$ADJUNCT
$Noun1
$Noun1
$Verb
$Verb
$Noun2?
$Noun2
d. $Noun1
this/that
je
$Noun1
$ADJUNCT
aux
$Verb
$ADJUNCT
$Noun2
$Verb $Noun2
(29) Lexicalizations
a. $Noun1
knjiga o Marku (book about Marko), politi?car (politician), predavanje
o komunizmu (lecture about communism), student (student), reklama
o nekretninama (advertisement about estate), snimatelj (cameraman),
govor ostipendiranju (speech aboutstipend), dekan(dean), istra?zivanje
o medijima (research about media), nau?cnik (scientist), pri?ca o Jovani
(story about Jovan), devojka (girl), predstava o Aleksandru (play about
Alexandar), koreograf (choreographer), mit o prosperitetu (myth about
prosperity), pisac (writer), ?sala o Peri (joke about Peter), devoj?cica
(girl), pesma o ?Zarku (song about Zarko), peva?c (singer), film o krim-
inalu (movie about criminal), momak (guy), ?clanak o adolescentima
(article about adolescents), psiholog (psychologist)
b. $Noun2
burne polemike (heated debate), knjiga o Marku (book about Marko),
neredi (riot), predavanje o komunizmu (lecture abut communism), reak-
cija (reaction), reklama o nekretninama (advertisement about estate),
situacija (situation), govor o stipendiranju (speech about stipend), re-
223
volt (revolt), istra?zivanje o medijima (research about media), odnos
(relation), pri?ca o Jovani (story about Jovana), problemi (problems),
predstava o Aleksandru (play about Alexandar), promene (changes),
mit o prosperitetu (myth about prosperity), napetost (tension), ?sala o
Peri (joke about Peter), pometnja (confusion), pesma o ?Zarku (pesma
about Zarko), pobuna (rebel), film o kriminalu (movie about criminal),
brojne kritike (numerous criticisms), ?clanak o adolescentima (article
about adolescents)
c. $ADJUNCT
pro?sle godine (last year), pro?slog meseca (last month), pro?sle nedelje
(last Sunday), pro?slog oktobra (last October), pro?slog ponedeljka (last
Monday), pro?slog utorka (last Tuesday), pro?sle srede (last Wednesday),
pro?slog ?cetvrtka (last Thrusday), pro?slog petka (last Friday), pro?sle
subote (last Saturday), pro?sle nedelje (last Sunday), pro?slog leta (last
summer)
d. $Verb
izazvati (evoke), pro?citati (read), zabele?ziti (writedown), snimiti (shoot),
olak?sati (make easier), odr?zati(hold), uraditi (do), zakomplikovati (com-
plicate), ispri?cati (tell), napraviti (cause), postaviti (put on stage),
doneti (bring), smanjiti (shrink), napraviti (make), pevati (sing), iz-
najmiti (rent), napisati (write)
224
Bibliography
Abels, Klaus. 2003. Successive cyclicity, anti-locality and adposition stranding.
Doctoral Dissertation, University of Connecticut.
Abels, Klaus, and Ad Neeleman. 2009. Universal 20 without the LCA. In Merging
features: computation, interpretation, acquisition., ed. J.M Brucart, A. Gavarro,
and J. Sola. Oxford University Press.
Aoun, Joseph, and Dominique Sportiche. 1983. On the formal theory of
government. The Linguistic Review 2:211?236.
Baker, Marc C. 1988. Incorporation: A theory of grammatical function changing.
Chicago: The University of Chicago Press.
Bard, Ellen Gurman, Dan Robertson, and Antonella Sorace. 1996. Magnitude
estimation of linguistic acceptability. Language 72:32?68.
Ba?si?c, Monika. 2004. Nominal subextractions and the structure of NPs in Serbian
and English. Master?s thesis, University of Tromso, Norway.
Bayer, Josef, Markus Bader, and Michael Meng. 2001. Morphological
underspecification meets oblique case: Syntactic and processing effects in
German. Lingua 111:465?514.
Behagel, Otto. 1923. Deutsche Syntax. Band I.-IV. Heidelberg.
den Besten, Hans. 1977/1983. On the interaction of root transformations and
lexical deletive rules. In On the formal syntax of the Westgermania, ed.
W. Abraham. Amsterdam: John Benjamins.
den Besten, Hans. 1985. The ergative hypothesis and free word order in Dutch and
German. In Studies in German grammar, ed. Jindrich Toman, 23?64.
Dordrecht: Foris.
Bobaljik, Jonathan, and Dianne Jonas. 1996. Subject positions and the rules of
TP. Linguistic Inquiry 27.2:195?236.
Bo?skovi?c, ?Zeljko. 2005. On the locality of left branch extraction and the structure
of NP. Studia Linguistica 59(1):1?45.
Broekhuis, Hans. 2005. Extraction from subjects: Some remarks on Chomsky?s On
phases. In Organizing grammar. Linguistic studies in honor of Henk van
Riemsdijk, ed. Hans Broekhuis et al., 59?68. Berlin/ New York: Mouton de
Gruyter.
Burge, Tyler. 1973. Reference and proper names. Journal of Philosophy
70:425?429.
225
B?uring, Daniel, and Katharina Hartmann. 1997. Doing the right thing. The
Linguistic Review 14:1?42.
Burzio, Luigi. 1981. Intransitive verbs and Italian auxiliaries. Doctoral
Dissertation, Massachusetts Institute of Technology.
Cattell, Ray. 1976. Constraints on movement rules. Language 52:18?50.
?Cavar, Damir, and Gisbert Fanselow. 2002. Distributed deletion. In Theoretical
approaches to universals, ed. A. Alexiadou. Amsterdam: John Benjamins.
Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Mass: MIT
Press.
Chomsky, Noam. 1973. Conditions on transformations. In A festschrift for Morris
Halle, ed. S. Anderson and P. Kiparsky, 232?286. New York: Holt, Rinehart &
Wilson.
Chomsky, Noam. 1975. Reflections on language. New York: Pantheon.
Chomsky, Noam. 1980. On binding. Linguistic Inquiry 11:1?46.
Chomsky, Noam. 1981. Lectures on government and binding: The Pisa lectures.
Mouton de Gruyter.
Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries. In Step by Step: Essays on
minimalist syntax in honor of Howard Lasnik, ed. R. Martin, D.Michaels, and
J. Uriagereka, 89?151. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2008. On phases. In Foundational issues in linguistic theory:
Essays in honor of Jean-Roger Vergnaud., ed. R. Freidin, C. Otero, and M.-L.
Zubzarreta. Cambridge, Mass.: MIT Press.
Cinque, Guglielmo. 1990. Types of A?-dependencies. Cambridge, Mass.: MIT
Press.
Corver, Norbert. 1991. The internal syntax and movement behavior of the Dutch
?wat voor?-construction. Linguistische Berichte 133:190?228.
Corver, Norbert. 1992. Left branch extraction. In Proceedings of NELS, 6784.
Corver, Norbert. 2006. Freezing effects. In The Blackwell Companion to Syntax
(Volume II), ed. M. Everaert and H. van Riemsdijk, 383?406. Malden, MA:
Backwell.
226
Culbertson, Jennifer, and Steven Gross. 2009. Are linguists better subjects? The
British journal for the philosophy of science 60:721?736.
Diesing, Molly. 1990. The syntactic roots of semantic partition. Doctoral
Dissertation, University of Massachusetts, Amherst, Ma.
Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press.
den Dikken, Marcel, Judy Bernstein, Christina Tortora, and Raffaella Zanuttini.
2007. Data and grammar: Means and individuals. Theoretical Linguistics
33(3):335?352.
Drummond, Alex. 2009. The unity of extraposition and the A/A-bar distinction.
In Proceedings of the 45th Annueal Meeting of the Chicago Linguistics Society.
Ernst, Thomas. 1984. Towards an integrated theory of adverb positions in English.
Doctoral Dissertation, University of Indiana, Bloomington.
Fanselow, Gisbert, Matthias Schlesewsky, and Ralf Vogel. 2008. Animacy effects
on crossing wh-movement in german. Unpublished manuscript, Universit?at
Potsdam.
Featherston, Sam. 2002. Coreferential objects in German: Experimental evidence
on reflexivity. Linguistische Berichte 192:457?484.
Featherston, Sam. 2005a. Magnitude estimation and what it can do for your
syntax: Some wh-constraints in German. Lingua 115:1525?1550.
Featherston, Sam. 2005b. That-trace in German. Lingua 115:1277?1302.
Featherston, Sam. 2007. Data in generative grammar: The stick and the carrot.
Theoretical Linguistics 33.3:269?318.
Featherston, Sam. to appear. Relax, lean back, and be a linguist. Zeitschrift f?ur
Sprachwissenschaft zum Thema ?Daten in der Sprachwissenschaft? .
Fox, Danny. 1995. Economy and scope. Natural Language Semantics 3:283?341.
Fujii, Tomohiro. 2006. Some theoretical issues in Japanese control. Doctoral
Dissertation, University of Maryland, College Park, Md.
Fukui, Naoki, and Yuji Takano. 1998. Symmetry in Syntax: Merge and Demerge.
Journal of East Asian Linguistics 7:27?86.
Gibson, Edward. 1998. Linguistic complexity: locality of syntactic dependencies.
Cognition 68:1?76.
Gibson, Edward, Neil Pearlmutter, Enriqueta Canseco-Gonzalez, and Gregory
Hickok. 1996. Recency preference in the human sentence processing mechanism.
Cognition 59:2359.
227
Goldberg, Adele. 2006. Constructions at work. Oxford University Press.
Goodall, Grant, Shinichiro Fukuda, Dan Michel, and Henry Beecher. 2010.
Comparing three methods for sentence judgment experiments. Presentation at
LSA Annual Meeting, Baltimore, Md.
Grewendorf, G?unther. 1988. Aspekte der Deutschen Syntax. Eine
Rektions-Bindungs-Analyse. T?ubingen: Narr.
Grewendorf, G?unther. 1989. Ergativity in German. Dordrecht: Foris.
Grewendorf, G?unther. 2002. Minimalistische Syntax. A. Francke Verlag T?ubingen
und Basel: UTB f?ur Wissenschaft.
Haider, Hubert. 1983. Connectedness effects in German. Groninger Arbeiten zur
Germanistischen Linguistik 25:83?117.
Haider, Hubert. 1993. Deutsche Syntax-generativ. T?ubingen: Narr.
Haider, Hubert. 2009. Linguistics - postmodern frippery or primordial cognitive
science. Presentation at the ?Visions for Linguistics? workshop, Univ. Konstanz,
Schloss Freudental.
Hankamer, Jorge, and Vera Lee-Schoenfeld. 2005. What moves in German
VP-fronting? Handout of presentation at Berkeley Syntax/Semantics Circle.
H?ohle, Tilman. 1982. Explikationen f?ur ?normale Betonung? und ?normale
Wortstellung?. In Satzglieder im Deutschen. Vorschl?age zur syntaktischen,
semantischen und pragmatischen Fundierung., ed. W. Abraham. T?ubingen:
Narr.
de Hoop, Helen. 1996. Case configuration and noun phrase interpretation.
Routledge.
Hornstein, Norbert, and Jairo Nunes. 2002. On asymmetries between Parasitic
Gap and Across-the-Board constructions. Syntax 5:1:26?54.
Huang, Jim. 1982. Logical relations in chinese and the theory of grammar.
Doctoral Dissertation, Massachussets Institute of Technology, Cambridge, Mass.
Hunter, Tim. 2010. The syntax and semantics of movement and adjunction in
natural language. Doctoral Dissertation, University of Maryland, College Park,
Md.
Ishii, Toru. 1997. An asymmetry in the composition of phrase structure and its
consequences. Doctoral Dissertation, University of California, Irvine.
Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, Mass.: MIT
Press.
228
Keller, Frank. 2000. Gradience in grammar: Experimental and computational
aspects of degress of grammaticality. Doctoral Dissertation, University of
Edinburgh.
Kikuchi, Akira. 1987. Comparative deletion in japanese. Unpublished manuscript,
Yamagata University.
Kitagawa, Yoshihisa. 1986. Subjects in Japanese and English. Doctoral
Dissertation, University of Massachusetts, Amherst, Ma.
Koopman, Hilda, and Dominique Sportiche. 1985. Theta theory and extraction.
Abstract in GLOW Newsletter 14.
Koopman, Hilda, and Dominique Sportiche. 1991. The position of subjecs. Lingua
85:211?258.
Kratzer, Angelika. 1996. Severing the external argument from its verb. In Phrase
structure and the lexicon, ed. J. Rooryck and L. Zaring, 109?137. Dordrecht,
Kluwer.
Kravtchenko, Ekaterina, Maria Polinsky, and Ming Xiang. 2009. Are all subject
islands created equal? Poster presented at CUNY 2009.
Larson, Richard K. 1988. On the double object construction. Linguistic Inquiry
19:335?391.
Lasnik, Howard, and Mamuro Saito. 1991. On the subject of infinitives. In Papers
from the 27th Regional Meeting of the Chicago Linguistic Society, ed. L. Dobrin,
L. Nichols, and R. Rodriguez, 324?343.
Lasnik, Howard, and Mamuro Saito. 1992. Move Alpha: Conditions on its
application and output. Cambridge, Mass.: MIT Press.
Lenerz, J?urgen. 1977. Zur Abfolge nominaler Satzglieder im Deutschen. T?ubingen:
Niemeyer.
Leu, Thomas. 2008. ?What for? internally. Syntax 11.1:1?25.
Levine, Robert, and Ivan A Sag. 2003. Some empirical issues in the grammar of
extraction. Proceedings of the HPSG03 conference .
Lidz, Jeff, and Alexander Williams. 2009. Constructions on holiday. Cognitive
Linguistics 20-1:177?189.
Longobardi, Giuseppe. 1994. Reference and proper names. Linguistic Inquiry
25:609?65.
Lutz, Ulrich. 2001. Studien zu Extraktion und Projektion im Deutschen. Doctoral
Dissertation, Universit?at T?ubingen.
229
Marantz, Alex. 1984. On the nature of grammatical relations. Cambridge MA:
MIT Press.
McCloskey, Jim. 1997. Subjecthood and subject positions. In Elements of
Grammar, ed. Liliane Haegeman, 197?235. Dordrecht: Kluwer.
Melchiors, Maria. 2007. What-for-splits and the analysis of dative objects in
German. Slides of the talk at Workshop on Object Positions, January 18-19,
2007. Aarhus Universitet, Aarhus, Denmark.
M?uller, Gereon. 1995. A-bar Syntax. Berlin: Mouton de Gruyter.
M?uller, Gereon. 1999. Optimality, markedness, and word order in German.
Linguistics 37:777?818.
M?uller, Gereon. 2004. Verb-second as vP-first. Journal of Comparative Germanic
Linguistics 7:179?234.
M?uller, Gereon. 2010. On deriving CED effects from the PIC. Linguistic Inquiry
41(1):35?82.
Murphy, Brian, and Carl Vogel. 2008. An empirical comparison of measurement
scales for judgements of acceptability. Poster presented at the Linguistic
Evidence Conference 2008, T?ubingen.
Nunes, Jairo. 2001. Sideward Movement. Linguistic Inquiry 32:2:303?344.
Nunes, Jairo. 2004. Linearization of chains and sideward movement. Linguistic
Inquiry Monograph. MIT Press.
Nunes, Jairo, and Juan Uriagereka. 2000. Cyclicity and extraction domains.
Syntax 3:20?43.
Pafel, J?urgen. 1996. Die syntaktische und semantische Struktur von was
f?ur-Phrasen. Linguistische Berichte 161:37?67.
Pereltsvaig, Asya. 2006. LBE negationslash= E: Split phrases in colloquial Russian. Handout of
presentation at FASL 15, University of Toronto.
Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In The
representation of (In)definiteness, ed. Eric Reuland Aliceter Meulen.
Cambridge, Mass.: MIT Press.
Phillips, Colin. 2009. Should we impeach armchair linguists? In Japanese/Korean
Linguistics 17, ed. S. Iwasaki et al. CSLI Publications.
Phillps, Colin, Matthew W. Wagers, and Ellen F. Lau. to appear. Grammatical
illusions and selective fallibility in real time language comprehension. Language
& Linguistics Compass .
230
Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure
of IP. Linguistic Inquiry 20.3:365?424.
Reinhart, Tanya. 1981. Definite NP anaphora and c-command. Linguisitic Inquiry
14:101?36.
Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry
24:657?720.
Richards, Marc. 2004. Object shift and scrambling in North and West Germanic:
A case study in symmetrical syntax. Doctoral Dissertation, University of
Cambridge.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of
grammar. Handbook of generative syntax, 281?337. Kluwer, Dordrecht.
Rohrbacher, Bernhard. 1994. Notes on the Antisymmetry of Syntax. In University
of Pennsylvania Working Papers in Linguistics, volume 1, 111?128.
Rosengren, Inger. 2002. EPP: A syntactic device in the servive of semantics.
Studia Linguistica 56(2):145?190.
Ross, John R. 1967. Constraints on variables in syntax. Doctoral Dissertation,
MIT.
Ross, John R. 1974. Three batons for cognitive psychology. In Cognition and the
symbolic processes, ed. W. B. Weimer and D. Palermo. Hillsdale, NJ: Lawrence
Erlbaum.
Sabel, Joachim. 1999. Das Passiv im Deutschen: Derivationale ?Okonomie vs.
Optionale Bewegung. Linguistische Berichte 177:87?112.
Saito, Mamuro, and Naoki Fukui. 1998. Order in phrase structure and movement.
Linguisitic Inquiry 29.3:439?474.
Sauerland, Uli, and Paul Elbourne. 2002. Total reconstruction, PF movement, and
derivational order. Linguistic Inquiry 33.2:283?319.
Sch?utze, Carson. 1996. The empirical base of linguistics: Grammaticality
judgments and linguistic methodology. Chicago: The University of Chicago
Press.
Sheehan, Michelle. 2009. Complement stranding: a window on multiple spellout
and the LCA. Presentation at GLOW 2009.
Sorace, Antonella. 2000. Gradients in auxiliary selection with intransitive verbs.
Language 76.4:859?890.
Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for
constituent structure. Lingusitic Inquiry 19:425?499.
231
Sprouse, Jon. 2007. A program for experimental syntax: Finding the relationship
between acceptability and grammatical knowledge. Doctoral Dissertation,
University of Maryland, College Park, Md.
Stepanov, Arthur. 2001. Cyclic domains in syntactic theory. Doctoral Dissertation,
University of Connecticut.
Stepanov, Arthur. 2007. The end of the CED? Minimalism and extraction
domains. Syntax 10:1:80?126.
Sternefeld, Wolfgang. 1985. Deutsch ohne grammatische Funktionen: ein Beitrag
zur Rektions-und Bindungstheorie. Linguistische Berichte 99:394?439.
Svenonius, Peter. 2001. Subject positions and the placement of adverbials. In
Subjects, Expletives and the EPP, ed. P. Svenonius. Oxford University Press,
New York.
Takahashi, Daiko. 1994. Minimality of movement. Doctoral Dissertation,
University of Connecticut.
Truswell, Robert. 2007. Extraction from adjuncts and the structure of events.
Lingua 117:1355?1377.
Uriagereka, Juan. 1999. Multiple spell-out. In Working minimalism, ed. S. Epstein
and N. Hornstein, 251?282. MIT Press.
Webelhuth, Gerd. 1989. Syntactic saturation phenomena and the modern
Germanic languages. Universit of Massacheusetts, GLSA.
Wellwood, Alexis, Roumyana Pancheva, Valentine Hacquard, Scott Fults, and
Colin Phillips. 2009. The role of event comparison in comparative illusions.
Poster presented at CUNY 2009.
Weskott, Thomas, and Gisbert Fanselow. 2008. Variance and informativity in
different measures of linguistic acceptability. In Proceedings of the 27th West
Coast Conference on Formal Linguistics, ed. Natasha Abner and Jason Bishop,
431?439.
Wexler, Ken, and Peter Culicover. 1981. Formal principles of language
acquisition.. Cambridge, Mass.: MIT Press.
Williams, Christopher John Fardo. 1985. Aristotle?s theory of descriptions.
Philosophical Review 94(1):63?80.
Williams, Edwin. 1981. Argument structure and morphology. Linguistic Review
1:81?114.
Wurmbrand, Susanne. 2004. No TP-fronting meets Nearly Headless Nick.
Unpublished manuscript, University of Connecticut.
232
Wurmbrand, Susanne. 2006. Licensing case. Journal of Germanic Linguistics
18.3:175?236.
Yoshida, Masaya. 2006. Constraints and mechanisms in long-distance dependency
formation. Doctoral Dissertation, University of Maryland, College Park, Md.
233