German Flemish

Incremental Processing Difficulty in
Cross-serial and Nested Verb Clusters
Flemish
Greg Kobele
Evelyne Lagrou
Felix Engelmann
a
V fin
De doctoraatstudent | heeft | een lezing | gehouden | tijdens het congres.
The Ph.D. student gave a talk during the conference
b
V-1
V fin
De professor | heeft | de doctoraatstudent | een lezing | laten | houden | tijdens het congres.
The professor let the Ph.D. student give a talk during the conference
c
subject helper
spillover
NP1
NP2
NP3
V-2
V-1
V fin
De leerling | heeft | de professor | de doctoraatstudent | een lezing | horen | laten | houden | tijdens het congres.
The student heard the professor let the Ph.D. student give a talk during the conference
d
V-3
V-2
V-1
V fin
De rector | heeft | de leerling | de professor | de doctoraatstudent | een lezing | willen | horen | laten | houden | tijdens het congres.
The director of the institute wanted the student to hear the professor let the Ph.D. student give a talk during the conference
German
Titus von der Malsburg
Ryan Musa
Sabrina Gerth
Ruben van de Vijver
John Hale
a
V fin
Der Doktorand | hat | einen Vortrag | gehalten | während der Konferenz.
The Ph.D. student gave a talk during the conference
b
V-1
V fin
Der Professor | hat | den Doktoranden | einen Vortrag | halten | lassen | während der Konferenz.
The professor let the Ph.D. student give a talk during the conference
Embedding verbs
Flemish German gloss
subject
helper
spillover
NP1
NP2
V-2
V-1
V fin
NP3
laten
lassen
to
let
Der Student | hat | den Professor | den Doktoranden | einen Vortrag | halten | lassen | gehört | während der Konferenz.
zien sehen
to see
The student heard the professor let the Ph.D. student give a talk during the conference
horen hören
to hear
willen wollen
to want
V-3
V-2
V-1
V fin
c
d
Der Institutsleiter | hat | den Studenten | den Professor | den Doktoranden | einen Vortrag | halten | lassen | hören | wollen | während der Konferenz.
The director of the institute wanted the student to hear the professor let the Ph.D. student give a talk during the conference
First Pass Reading Time (level d)
Total Fixation Time
←Observation 1
Much slower beyond level b.
●
●
●
600
Flemish
●
●
●
●
●
●
●
●
500
●
Replicates Bach, Brown & Marslen-Wilson, 1986
Mean TFT in ms
●
400
●
●
●
●
●
●
●
●
cond
a
b
c
d
●
300
●
Observation 2→
Along the verb cluster,
German readers slow down.
Flemish readers don’t.
●
●
800
●
●
700
●
●
●
●
●
●
●
●
400
●
●
●
●
●
●
●
●
●
300
German
600
500
340
●
●
●
●
subj+hlpr
objects
V−3
V−2
V−1
Region
V_fin
Mean FPRT in ms
800
700
360
●
320
●
300
●
280
●
●
●
●
●
260
240
V−3
Account of Observation 2
A top-down parser confirms Flemish verbal expectations in order
of their prediction, whereas German predictions are confirmed in
reverse order. This selectively burdens German readers’ memory.
d= d= v
v= d= v
=v i
Derivation trees for level c
1
Flemish
German
confirmation order is same as [prediction] order
confirmation order is reverse of [prediction] order
0
merge1
1
helper2
3
[5]
1
merge3
3
merge5
5
V-113
V_fin
spillover
Minimalist grammar
0
V−1
Region
language
transitive verb
embedding verb
helper
V−2
1
helper2
3
subject4
[5]
merge6
merge1
←Verbs raise into the cluster.
Their order reflects languageparticular linearization.
3
5
German
merge3
merge5
V fin15
Flemish
subject4
merge6
Memory use in incremental parser states
[11]
V fin15
11
NP312
9
9
NP210
[11]
merge11
V-213
11
20
merge9
NP312
9
NP210
For more on stack tenure, see Kobele et al in Proceedings of Formal Grammar,
2012. For more on top-down parsing and Minimalist Grammars, see Stabler in
Proceedings of the ACL Cognitive Modeling and Computational Linguistics
Workshop, 2011. For more on the raising analysis of verb clusters as formalized
in Minimalist Grammars, see chapter 9 of Morawietz, 2003.
Flemish
German
15
merge11
8
V-114
NP17
10
9
[8]
merge9
6
merge8
5
8
V-214
6
NP17
0
[8]
6
merge8
sum stack tenure
6
−4
−3
−2
−1
steps from finishing
0