```Working Memory and Relative Clause Attachment under
Increased Sentence Complexity
Akira Omaki
Department of Second Language Studies, University of Hawai‘i at Mānoa
1. Introduction
3. Experiment 1
Method
Relative Clause Attachment
Local
Attachment
(1) Someone shot…
…the servant of the actress [who was on the balcony].
(N1)
(N2)
(RC)
Results
Figure 1. M ean NLA responses & RST scores in Experiment 1
 Participants: 36 English native speakers
 Materials: 32 target items (4) + 75 fillers
Mean NLA: SC
(4a) Embedded Condition (EC)
burned herself the other day adored]] was very nice.
Local Attachment (LA)
English, Arabic, Norwegian, Romanian, Croatian, etc.
Non-local Attachment (NLA)
Spanish, Dutch, Afrikaans, French, Greek, Russian etc.
Various Accounts for the cross-linguistic differences:
• Implicit Prosody (Fodor, 2002)
• Construal Theory (Frazier & Clifton, 1996)
• Tuning Hypothesis (Mitchell et al., 1995)
• Attachment Binding (Hemforth et al., 1998)
The babysitter said [that the sister of the schoolgirl
who burned herself the other day was very nice.]
 Mendelsohn & Pearlmutter (1999):
English object-modifying RC + Reading Span Test
 Swets, Desmet, Hambrick & Ferreira (2004):
Subject modifying RC in English (LA preference) and
Dutch (NLA preference) + Reading Span Test
20%
0%
20.00
Example Question:
Who got burned the other day? 1. the sister 2. the
schoolgirl
30.00
40.00
50.00
60.00
70.00
RST scores
• Significant negative correlation btw. RST scores and NLA responses
2-tailed correlation in…
EC condition: r = - .482 (p<.001) SC condition: r = - .561 (p<.001)
 The greater RS leads to an LA preference. (Replicates previous findings)
1. Read the whole sentence on a computer screen
2. then press the space bar to show the question
Offline experiments by…
40%
 Less complex (Storage cost at “who” = 2)
A. Offline reading experiment on computer
Individual differences in RC attachment preferences?
Working Memory Capacity
60%
(4b) Sentence-complement Condition (SC)
 Procedure
HOWEVER…
Trend line (SC)
80%
[The babysitter [that the sister of the schoolgirl who
Cross-linguistic differences in RC attachment preferences:
Trend line (EC)
100%
 More complex (Storage cost at “who” = 4)
Non-local
Attachment
Mean NLA: EC
• No significant difference between EC & SC condition
(Correlation between EC and SC condition was over .90)
B. Reading Span Test: A version of Waters &
Caplan’s (1996)
• Acceptability judgment + Recall Task
• Score 1 was added when both judgment and recall were
correct (Max score: 70)
• Weak trend for a local attachment preference
Mean NLA responses in…
EC condition: 42.19% (SD=27.60) SC condition: 46.35% (SD=28.48)
Question: Why the lack of sentence complexity effect?
A. Problems with the methodology? They may not have read the whole sentence, since one can answer the question by just
Their Findings:
The greater WM capacity leads to an LA preference
(i.e., Low-spans prefer NLA, High-spans prefer LA)
4. Experiment 2 (in progress)
Method
 Very surprising, in that various locality principles are
supposed to minimize processing burden (e.g., Late
Closure (Frazier, 1987), Recency (Gibson et al., 1996))
2. Goal of the Present Study
Preliminary RT Results
 Main effect of complexity (F1&F2) in ‘of’ ‘schoolgirl’ ‘who’ burned’
 Reflects the storage cost (Gibson, 2000)
 Participants: 32 English native speakers
 Materials: 32 target items (5) + 73 fillers
 Procedure: Self-paced reading + RST from Exp 1
(5a, c) Complex/Non-complex, forced LA
The babysitter (said) that the brother of the schoolgirl
who burned herself the other day adored was very nice.
(5b, d) Complex/Non-complex, forced NLA
The babysitter (said) that the brother of the schoolgirl
who burned himself the other day was very nice.
To further examine whether WM capacity
interacts with the RC attachment ambiguity in
off-line processing (Exp 1 ) and on-line
processing (Exp 2) by investigating…
 Main effect of complexity in ‘was’ (F1&F2)
 Reflects the integration cost (Gibson, 2000)
 Main effect of attachment (though only F1) in ‘himself/herself’
 LA attachment preference, regardless of the sentence complexity
 No main effect of interaction in any of the regions (F1&F2)
Figure 2. Overall Reaction Times by region in Experiment 2
(a) whether high-spans and low-spans differ
in RC attachment preferences (as in
previous studies), and
Complex/LA
Complex/NLA
Non-complex/LA
Non-Complex/NLA
Storage cost
1600
(b) whether the sentence complexity (i.e.,
the amount of memory cost) interacts
with RC attachment preferences
(cf. Eastwick & Phillips, 1999)
1400
1200
LA preference
Integration
cost
1000
800
600
High vs. Low: Between-subjects
400
The
babysitter
(said)
• Mean RST score (n = 32): 45.9, thus 46 as cut-off
 16 low-spans, 16 high-spans
 Complexity (2) x attachment (2) x span size (2) design
1150
900
650
the
brother
1. High-spans generally slower
2. Integration, but NOT attachment,
interacts with span size
Regions with main effect of span size: ‘himself/herself’ and ‘was’
Figure 3: Span size effect in Mean RTs at 'himself/herself'
that
Figure 4: Span size effect in Mean RTs at 'was'
1200
Main Effect: RST p<.05,
Attach p=.053
No interaction
1023.68
925.18 906.74
1000
802.29
634.41
800
600
537.24 552.35
541.71
Main Effect: RST p<.01,
Complex p<.0005
Complx*RST p<.05
955.48
860.26
584.48 580.20
503.06 467.73
445.38
Low Span
Complex/LA
Complex/NLA
High Span
Non-Complex/LA
Non-Complex/NLA
the
schoolgirl
who
burned
himself/
herself
the other day
was
5. Summary
• The general LA preference was observed in offline and online processing
• RS significantly correlated with RC attachment ONLY in offline processing.
In Exp 2, RS significantly interacted with integration cost, but not with
attachment (cf. Caplan & Waters, 1999) – this suggests that (at least) initial RC
attachment preference is not influenced at all by memory capacity.
402.19
6. Acknowledgement
400
400
of
200
Low Span
Complex/LA
Complex/NLA
High Span
Non-Complex/LA
Non-Complex/NLA
• I thank Matt Prior, Amy J. Schafer, Barbara Schulz, and Bonnie D. Schwartz for their help at various stages. I
also thank the snack-eaters at the Psycholinguistics 3ji-no Oyatsu meetings.
• This research was partially supported by Elizabeth Holmes-Carr Scholarship, received with much appreciation.
```