Competing Embedded Clauses in German

Competing Embedded Clauses in German
Matthias Schrinner
Data
Clause-final relative clauses (RC) and argument clauses (AC) may be serialized RC»AC or AC»RC. Haider (1994) notices asymmetry in grammaticality;
claims that AC»RC is ungrammatical (Haider Asymmetry).
(1)
a.
Es fiel letzte Woche einem Grammatiker auf [RC der das untersuchte ] [AC dass dieser Satz grammatisch ist ]
It struck a grammarian last week, who analyzed it, that this clause is grammatical.
Es fiel letzte Woche einem Grammatiker auf [AC dass dieser Satz grammatisch ist ] [RC der das untersuchte]
b.
Claim
Constraints on Serialization
Both serializations are grammatical.
Haider Asymmetry persists in preference, not in grammaticality.
When RC is extraposed out of subject and AC is object, there is no
preference (Anti-Haider-Asymmetry-Context; AHA-Context).
Relevant Principles for Word Order
Principle of End-Recursion: Final-embedding structures are preferred
over center-embedding structures. (Miller & Chomsky, 1963)
Principle of Proximity: There is proximity between governors and
dependents. There are no crossing dependencies.
(Stowell, 1981; Ferrer i Cancho, 2006)
Investigation
Motivation: unclear empirical situation
Haider (1994) notices asymmetry
Keller (1994), concerned with extraposition in general, does not notice asymmetry
Constraints
Experiment: questionnaire
51 German native speakers were asked to rate sentences with extraposed AC and RC
as well as (clearly) grammatical and ungrammatical filler sentences.
Grammatical functions of clauses were permuted.
Scale from 1 (best) to 6 (worst)
(2)
a.
b.
Results
None of the structures is ungrammatical.
In general, RC»AC is preferred.
There is no Haider Asymmetry if RC antecedent is subject and AC is direct object.
c.
AHA-Context Haider Asymmetry Haider Asymmetry
RC antecedent SUB
AC DO
DO
SUB
IO
DO
total
fillers
RC»AC 3.25
AC»RC 3.17
2.78
3.56
3.17
3.84
3.07
3.61
gramm. 1.95
ungramm. 5.15
d.
E XTRAPOSITION (E XTRAP )
CPs with a [+extrap]-feature are extraposed.
high ranked; eliminates non-extraposed clauses
M ODIFIER A DJACENCY (MA)
Assign one * per linearly intervening projection between a modifier and its
licenser.
favors RC»AC
A RGUMENT A DJACENCY (AA)
Assign one * per linearly intervening projection between an argument and
its licenser.
favors AC»RC
*C ROSSING D EPENDENCIES (*C R D EP )
Surface structures do not exhibit crossing dependencies.
Ordering MA over AA (MA»AA) will favor RC»AC serializations. Making use of
Stochastic OT will allow AC»RC as well. Adding a lower ranked *C R D EP explains why in
AHA-Context there is no preference of RC»AC (see below).
Analysis in Stochastic Optimality Theory
M ODIFIER A DJACENCY
Assumptions
Everything is base generated under VP.
(Haider, 1993)
Extraposition is rightward Chomsky-Adjunction to VP.
(Guéron, 1980; Sternefeld & von Stechow, 1988;
Grewendorf, 1988; Büring & Hartmann, 1995)
Stochastic OT: Every constraint can be top ranked.
(Boersma & Hayes, 2001)
VP
DP
V
Haider Asymmetry
VP
VP
VP
Argument Adjacency
AA
(5) RC»AC *!
☞ (6) AC»RC
CP-relj
CP-argi
V′
DP
RC»AC
Es ist dem Grammatiker aufgefallen [CP der es
untersuchte ] dass der Satz gut ist
AC»RC
Es ist dem Grammatiker aufgefallen dass der
Satz gut ist der es untersuchte
MA
☞ (3) RC»AC *
(4) AC»RC **!
V′
ej
RC»AC
Es ist dem Grammatiker [V aufgefallen ] der es
untersuchte dass der Satz gut ist
AC»RC
Es ist dem Grammatiker [V aufgefallen ]
[CP dass der Satz gut ist ] der es untersuchte
ei
(6)
CPi [+extrap]
VP
(4)
(5)
CPj [+extrap]
VP
(3)
*C ROSSING D EPENDENCIES
AC»RC exhibits crossing
dependencies.
*C R D EP strengthens effect of
MA»AA.
V
N . . . ej
AHA-Context
VP
VP
VP
CP-argj
CP-reli
DP
V′
N . . . ei
ej V
RC»AC exhibits crossing
dependencies.
*C R D EP weakens effect of MA»AA.
N . . . ei
Ranking
References
In Stochastic OT, the evaluation is carried out making use of selection points, which are
normally distributed around a mean value (as given in the figure). At every utterance,
new selection points are chosen and a new competition is run. The results of the
competitions are equal to the judgements of the speakers.
M ODIFIER A DJACENCY
A RGUMENT A DJACENCY
*C ROSSING D EPENDENCIES
speakers’ ratings 1 or 2
Haider Asymmetry RC»AC 415
AC»RC 207
AHA-Context
RC»AC
AC»RC
80
98
simulation
Boersma, Paul & Bruce Hayes. 2001. Empirical Tests of the Gradual Learning Algorithm. Linguistic Inquiry 45–86.
Büring, Daniel & Katharina Hartmann. 1995. All right! In U. Lutz & J. Pafel (eds.), On Extraction and Extraposition in German,
179–212.
Ferrer i Cancho, Ramon. 2006. Why do syntactic links not cross? EuroPhysics Letters 76. 1228–1234.
Grewendorf, Günther. 1988. Aspekte der deutschen Syntax. Tübingen: Narr.
Guéron, Jacqueline. 1980. On the Syntax and Semantics of PP Extraposition. Linguistic Inquiry 637–678.
Haider, Hubert. 1993. Deutsche Syntax – Generativ. Tübingen: Narr.
Haider, Hubert. 1994. Detached Clauses—The Later The Deeper. Arbeitspapier SFB 340 No 41.
Keller, Frank. 1994. Extraposition in HPSG. Verbmobil Report No 30.
Miller & Noam Chomsky. 1963. Finitary models of language users. In R. D. Luce (ed.), Handbook of Mathematical Psychology,
269–321.
Sternefeld, Wolfgan & Arnim von Stechow. 1988. Bausteine syntaktischen Wissens. Opladen: Westdeutscher Verlag.
Stowell, Timothy A. 1981. Origins of Phrase Structure: MIT dissertation.
Haider Asymmetry RC»AC 663
AC»RC 337
AHA-Context
18
16
14
12
10
8
6
4
2
RC»AC 508
AC»RC 492
Conclusion
Serialization of embedded clauses in German is subject to preference.
RC»AC is the preferred structure.
In AHA-Context, there is no preference.
Modelling should make use of a stochastic model of grammar.
Acknowledgments
I would like to thank Gereon Müller, Sandra Döring, Fabian Heck, Jonathan Bobaljik,
Robert Truswell, and the anonymous commentators of ConSOLE and DGfS Annual
Meeting for their useful commentaries, questions and support. I would like to offer my
special thanks to Andreas Opitz for the assistance with the statistics in this project.