Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Readers with less cognitive control are moreaffected by surprising content: Evidence from a
self-paced reading experiment in German
Bruno Nicenboim, Shravan Vasishth and Reinhold Kliegl
University of Potsdam
1 / 23
Aim:
I investigate how the processing of expected and unexpectedupcoming words is affected by
I cognitive controlI working memory capacityI reading skills
2 / 23
Introduction
How to disentangle the easier integration into the context andpredictions?
a strong prediction for a noun phrase will trigger expectations ofcertain features in its previous determiner or adjective:
(1) DeLong, Urbach, and Kutas (2005)
a. The day was breezy so the boy went outside to fly a kite.
b. The day was breezy so the boy went outside to fly an airplane.
3 / 23
Introduction
How to disentangle the easier integration into the context andpredictions?
a strong prediction for a noun phrase will trigger expectations ofcertain features in its previous determiner or adjective:
(1) DeLong, Urbach, and Kutas (2005)
a. The day was breezy so the boy went outside to fly a kite. pre-dictable determiner
b. The day was breezy so the boy went outside to fly an airplane. un-predictable/surprising determiner
3 / 23
Study Lang. Method Results
DeLong et al. (2005) English written ERP N400
Van Berkum et al. (2005) Dutch audio ERP positive deflection
experiment 1 (and N400 at the unpredictable nouns)
Van Berkum et al. (2005) Dutch SPR effect at spillover
experiment 3
Otten et al. (2007) Dutch audio ERP right frontal electrodes negativity
Dutch × context (300-600 ms) in the predictive context
Otten and Van Berkum (2008) Dutch written ERP negativity (900-1100 ms)
Experiment 1B × context only in the predictive context
Otten and Van Berkum (2009) Dutch audio ERP early negative deflection (300–600 ms)
× wmc additional later negativity for low-wmc
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2004) Spanish written ERP distributed positivity (500–700 ms)
Szewczyk and Schriefers (2013) Polish written ERP a parieto-occipitally
distributed negativity (400–600 ms)
Table: List of studies
4 / 23
Study Lang. Method Results
DeLong et al. (2005) English written ERP N400
Van Berkum et al. (2005) Dutch audio ERP positive deflection
experiment 1 (and N400 at the unpredictable nouns)
Van Berkum et al. (2005) Dutch SPR effect at spillover
experiment 3
Otten et al. (2007) Dutch audio ERP right frontal electrodes negativity
Dutch × context (300-600 ms) in the predictive context
Otten and Van Berkum (2008) Dutch written ERP negativity (900-1100 ms)
Experiment 1B × context only in the predictive context
Otten and Van Berkum (2009) Dutch audio ERP early negative deflection (300–600 ms)
× wmc additional later negativity for low-wmc
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2004) Spanish written ERP distributed positivity (500–700 ms)
Szewczyk and Schriefers (2013) Polish written ERP a parieto-occipitally
distributed negativity (400–600 ms)
Table: List of studies
4 / 23
Study Lang. Method Results
DeLong et al. (2005) English written ERP N400
Van Berkum et al. (2005) Dutch audio ERP positive deflection
experiment 1 (and N400 at the unpredictable nouns)
Van Berkum et al. (2005) Dutch SPR effect at spillover
experiment 3
Otten et al. (2007) Dutch audio ERP right frontal electrodes negativity
Dutch × context (300-600 ms) in the predictive context
Otten and Van Berkum (2008) Dutch written ERP negativity (900-1100 ms)
Experiment 1B × context only in the predictive context
Otten and Van Berkum (2009) Dutch audio ERP early negative deflection (300–600 ms)
× wmc additional later negativity for low-wmc
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2004) Spanish written ERP distributed positivity (500–700 ms)
Szewczyk and Schriefers (2013) Polish written ERP a parieto-occipitally
distributed negativity (400–600 ms)
Table: List of studies
4 / 23
Study Lang. Method Results
DeLong et al. (2005) English written ERP N400
Van Berkum et al. (2005) Dutch audio ERP positive deflection
experiment 1 (and N400 at the unpredictable nouns)
Van Berkum et al. (2005) Dutch SPR effect at spillover
experiment 3
Otten et al. (2007) Dutch audio ERP right frontal electrodes negativity
Dutch × context (300-600 ms) in the predictive context
Otten and Van Berkum (2008) Dutch written ERP negativity (900-1100 ms)
Experiment 1B × context only in the predictive context
Otten and Van Berkum (2009) Dutch audio ERP early negative deflection (300–600 ms)
× wmc additional later negativity for low-wmc
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2004) Spanish written ERP distributed positivity (500–700 ms)
Szewczyk and Schriefers (2013) Polish written ERP a parieto-occipitally
distributed negativity (400–600 ms)
Table: List of studies
4 / 23
Study Lang. Method Results
DeLong et al. (2005) English written ERP N400
Van Berkum et al. (2005) Dutch audio ERP positive deflection
experiment 1 (and N400 at the unpredictable nouns)
Van Berkum et al. (2005) Dutch SPR effect at spillover
experiment 3
Otten et al. (2007) Dutch audio ERP right frontal electrodes negativity
Dutch × context (300-600 ms) in the predictive context
Otten and Van Berkum (2008) Dutch written ERP negativity (900-1100 ms)
Experiment 1B × context only in the predictive context
Otten and Van Berkum (2009) Dutch audio ERP early negative deflection (300–600 ms)
× wmc additional later negativity for low-wmc
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2003) Spanish written ERP N400
and drawings
Wicha et al. (2004) Spanish written ERP distributed positivity (500–700 ms)
Szewczyk and Schriefers (2013) Polish written ERP a parieto-occipitally
distributed negativity (400–600 ms)
Table: List of studies
4 / 23
Working memory capacity (WMC)Does the ability to temporarily store and manipulateinformation (WMC) affect the ability to predict upcomingwords?
We can measure WMC using the operation span task (Turner &Engle, 1989)
low mid highWorking Memory Capacity
RT
condition
predictable
unpredictable
but note the findings of Otten and Van Berkum (2009)
5 / 23
Working memory capacity (WMC)Does the ability to temporarily store and manipulateinformation (WMC) affect the ability to predict upcomingwords?We can measure WMC using the operation span task (Turner &Engle, 1989)
low mid highWorking Memory Capacity
RT
condition
predictable
unpredictable
but note the findings of Otten and Van Berkum (2009)
5 / 23
Working memory capacity (WMC)Does the ability to temporarily store and manipulateinformation (WMC) affect the ability to predict upcomingwords?We can measure WMC using the operation span task (Turner &Engle, 1989)
low mid highWorking Memory Capacity
RT
condition
predictable
unpredictable
but note the findings of Otten and Van Berkum (2009)5 / 23
Reading skillsDoes the “prediction ability” depends on the sensitivity to thesemantic cues characteristic of skilled readers (Pearlmutter &MacDonald, 1995)?
We can measure reading skills using the rapid automatized namingtask (Denckla & Rudel, 1976)
low mid highReading Skills
RT
condition
predictable
unpredictable
6 / 23
Reading skillsDoes the “prediction ability” depends on the sensitivity to thesemantic cues characteristic of skilled readers (Pearlmutter &MacDonald, 1995)?
We can measure reading skills using the rapid automatized namingtask (Denckla & Rudel, 1976)
low mid highReading Skills
RT
condition
predictable
unpredictable
6 / 23
Reading skillsDoes the “prediction ability” depends on the sensitivity to thesemantic cues characteristic of skilled readers (Pearlmutter &MacDonald, 1995)?
We can measure reading skills using the rapid automatized namingtask (Denckla & Rudel, 1976)
low mid highReading Skills
RT
condition
predictable
unpredictable
6 / 23
Cognitive Control
Cognitive control is related to: (Botvinick, Braver, Barch, &Carter, 2001)
I adjustments in perceptual selection
I response biasing
I on-line maintenance of contextual information
I tasks requiring the subject to override relativelyautomatic but task-inappropriate responses (Stroop)
Suppression ability
Ability to suppress irrelevant information during comprehensionindependent of the ability to enhance relevant information(Structure-Building Framework: Gernsbacher, 1997)
7 / 23
Cognitive Control
Cognitive control is related to: (Botvinick, Braver, Barch, &Carter, 2001)
I adjustments in perceptual selection
I response biasing
I on-line maintenance of contextual information
I tasks requiring the subject to override relativelyautomatic but task-inappropriate responses (Stroop)
Suppression ability
Ability to suppress irrelevant information during comprehensionindependent of the ability to enhance relevant information(Structure-Building Framework: Gernsbacher, 1997)
7 / 23
Cognitive Control
Cognitive control is related to: (Botvinick, Braver, Barch, &Carter, 2001)
I adjustments in perceptual selection
I response biasing
I on-line maintenance of contextual information
I tasks requiring the subject to override relativelyautomatic but task-inappropriate responses (Stroop)
Suppression ability
Ability to suppress irrelevant information during comprehensionindependent of the ability to enhance relevant information(Structure-Building Framework: Gernsbacher, 1997)
7 / 23
Cognitive Control
Cognitive control ( ≈ Suppression ability) in associative priming(Boudewyn, Long, & Swaab, 2012):
(2) a. In her haste she forgot to buy the apples and oranges.associated
b. In her haste she forgot to buy the apples and bread.unassociated
Results:
I stronger N400 effects in words unassociated (bread) to theprime in comparison with associated words (oranges)
I participants with less cognitive control showed stronger N400effects in the unassociated words (bread) in comparison withthe associated ones (oranges)
8 / 23
Cognitive Control
Cognitive control ( ≈ Suppression ability) in associative priming(Boudewyn et al., 2012):
(2) a. In her haste she forgot to buy the apples and oranges.associated
b. In her haste she forgot to buy the apples and bread.unassociated
Results:
I stronger N400 effects in words unassociated (bread) to theprime in comparison with associated words (oranges)
I participants with less cognitive control showed stronger N400effects in the unassociated words (bread) in comparison withthe associated ones (oranges)
8 / 23
Cognitive Control
Cognitive control ( ≈ Suppression ability) in associative priming(Boudewyn et al., 2012):
(2) a. In her haste she forgot to buy the apples and oranges.associated
b. In her haste she forgot to buy the apples and bread.unassociated
Results:
I stronger N400 effects in words unassociated (bread) to theprime in comparison with associated words (oranges)
I participants with less cognitive control showed stronger N400effects in the unassociated words (bread) in comparison withthe associated ones (oranges)
8 / 23
Cognitive Control
Cognitive control ( ≈ Suppression ability) in associative priming(Boudewyn et al., 2012):
(2) a. In her haste she forgot to buy the apples and oranges.associated
b. In her haste she forgot to buy the apples and bread.unassociated
Results:
I stronger N400 effects in words unassociated (bread) to theprime in comparison with associated words (oranges)
I participants with less cognitive control showed stronger N400effects in the unassociated words (bread) in comparison withthe associated ones (oranges)
8 / 23
Cognitive Control
How can we measure cognitive control?
Stroop task:
BLUE - RED
De Houwer’s (2003) variation of the Stroop task (Stroop, 1935)with also a neutral condition (Brown, 2011)
Cognitive Control
How can we measure cognitive control?
Stroop task: BLUE
- RED
De Houwer’s (2003) variation of the Stroop task (Stroop, 1935)with also a neutral condition (Brown, 2011)
Cognitive Control
How can we measure cognitive control?
Stroop task: BLUE - RED
De Houwer’s (2003) variation of the Stroop task (Stroop, 1935)with also a neutral condition (Brown, 2011)
Cognitive Control
How can we measure cognitive control?
Stroop task: BLUE - RED
De Houwer’s (2003) variation of the Stroop task (Stroop, 1935)with also a neutral condition (Brown, 2011)
Cognitive Control: Stroop taskCongruent condition:
+
10 / 23
Cognitive Control: Stroop taskCongruent condition:
BLUE
10 / 23
Cognitive Control: Stroop taskNeutral condition:
+
10 / 23
Cognitive Control: Stroop taskNeutral condition:
TABLE
10 / 23
Cognitive Control: Stroop taskSemantic interference condition:
+
10 / 23
Cognitive Control: Stroop taskSemantic interference condition:
BLUE
10 / 23
Cognitive Control: Stroop taskResponse interference condition:
+
10 / 23
Cognitive Control: Stroop taskResponse interference condition:
YELLOW
10 / 23
Cognitive Control: Stroop taskDe Houwer’s 2003 version:
“Subtractive” logic in the design:Congruent < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Congruent − Semantic interference
But according to Brown (2011), the Congruent conditionalso includes facilitation effects uncorrelated with theinterference effects:
“Subtractive” logic in the design:Congruent < Neutral < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Neutral − Semantic interference
11 / 23
Cognitive Control: Stroop taskDe Houwer’s 2003 version:
“Subtractive” logic in the design:Congruent < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Congruent − Semantic interference
But according to Brown (2011), the Congruent conditionalso includes facilitation effects uncorrelated with theinterference effects:
“Subtractive” logic in the design:Congruent < Neutral < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Neutral − Semantic interference
11 / 23
Cognitive Control: Stroop taskDe Houwer’s 2003 version:
“Subtractive” logic in the design:Congruent < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Congruent − Semantic interference
But according to Brown (2011), the Congruent conditionalso includes facilitation effects uncorrelated with theinterference effects:
“Subtractive” logic in the design:Congruent < Neutral < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Neutral − Semantic interference
11 / 23
Cognitive Control: Stroop taskDe Houwer’s 2003 version:
“Subtractive” logic in the design:Congruent < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Congruent − Semantic interference
But according to Brown (2011), the Congruent conditionalso includes facilitation effects uncorrelated with theinterference effects:
“Subtractive” logic in the design:Congruent < Neutral < Semantic interference < Response interference
Semantic Conflict as a measure of Cognitive Control(De Houwer, 2003):
Semantic Conflict = Neutral − Semantic interference
11 / 23
Cognitive Control
Is associative priming may be a special case of the generalmechanism used to make predictions?
If so:
low mid highCognitive Control
RT
condition
unpredictable
predictable
12 / 23
Cognitive Control
Is associative priming may be a special case of the generalmechanism used to make predictions?
If so:
low mid highCognitive Control
RT
condition
unpredictable
predictable
12 / 23
Aim:
I investigate how the processing of expected and unexpectedupcoming words is affected by
I cognitive controlI working memory capacityI reading skills
13 / 23
Experiment
Pretest
I Task: Sentence completion
I Subjects: 38
I Stimuli: 53 incomplete experimental sentences such as (3)and 57 fillers
(3) DerThe
Backerbaker
schobput
denthe
Teigdough
inin
...
...
26 sentences completed with nouns with cloze probability over 0.61where selected.
14 / 23
Stimuli
(4) a. predictable
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinenhis.masc
Ofenoven
undand
wartete.waited
b. unpredictable/surprising
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinehis.fem
Mikrowellemicrowave
undand
wartete.waited
15 / 23
Experiment
Experiment
I Subjects: 81I Tasks:
I assessment of individual differencesI Adaptation of De Houwer’s (2003) Stroop task for cognitive
controlI Operation span task (Conway et al., 2005; Malsburg &
Vasishth, 2012) for working memory capacityI Rapid automatized naming (Kuperman & Van Dyke, 2011) for
reading skills
I Self-paced readingI Stimuli: 26 items based on the pretest such as the previous
example (4) and fillers
16 / 23
Results of Stroop taskStroop conditions are significantly different from each otherand:Congruent < Neutral < Semantic interference < Response interference
●
●
●
●
●
●
●
●
510
520
530
540
550
560
C N SI RIStroop conditions
RT
(m
s)
Figure: Visualization of LMM partial effects (Hohenstein & Kliegl, 2013)17 / 23
Assessment of individual differences
I Cognitive control = weighted difference (using PCA) ofmean speed of Semantic interference and Neutral conditions
I Working Memory Capacity = partial credit units (Conwayet al., 2005)
I Reading Skills = mean speed of the rapid automatizednaming task
18 / 23
Results
I Numerical difference between the conditions at the determiner(Coef = 0.01, SE = 0.01, t = .81)
I Significant difference between the conditions at the noun in alength-corrected model (Coef = 0.03, SE = 0.02, t = 2.18)
●●
●●
●
●
●
●
350
370
390
410
...in
...inseinen/seine
his.masc/his.femDETERMINER
Ofen/Mikrowelleoven/microwave
NOUN
und...and...
Region
RT
(m
s)
condition ● ●predictable unpredictable
19 / 23
ResultsI Numerical difference between the conditions at the determiner
(Coef = 0.01, SE = 0.01, t = .81)
I Significant difference between the conditions at the noun in alength-corrected model (Coef = 0.03, SE = 0.02, t = 2.18)
●●
●●
●
●
●
●
350
370
390
410
...in
...inseinen/seine
his.masc/his.femDETERMINER
Ofen/Mikrowelleoven/microwave
NOUN
und...and...
Region
RT
(m
s)
condition ● ●predictable unpredictable
19 / 23
ResultsI no significant interactions between reading skills or working
memory and condition
I significant cognitive control × condition interaction (Coef =
-0.05, SE = 0.01, t = -3.61)
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●●
●
●
●●
●●
●●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
290
315
340
365
390
−2 −1 0 1 2Cognitive Control
RT
(m
s)
condition
●
●
predictable
unpredictable
Figure: Visualization of LMM partial effects (Hohenstein & Kliegl, 2013)
20 / 23
ResultsI no significant interactions between reading skills or working
memory and conditionI significant cognitive control × condition interaction (Coef =
-0.05, SE = 0.01, t = -3.61)
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●●
●
●
●●
●●
●●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
290
315
340
365
390
−2 −1 0 1 2Cognitive Control
RT
(m
s)
condition
●
●
predictable
unpredictable
Figure: Visualization of LMM partial effects (Hohenstein & Kliegl, 2013)20 / 23
Conclusions
I No significant interactions between reading skills or workingmemory and condition, probably because the experimentalsentences were too short and simple
I Longer RTs at the unpredictable determiner (numerically) andat the noun (significant), which is consistent with the resultsfrom the literature (Van Berkum, Brown, Zwitserlood,Kooijman, & Hagoort, 2005)
21 / 23
Conclusions
I No significant interactions between reading skills or workingmemory and condition, probably because the experimentalsentences were too short and simple
I Longer RTs at the unpredictable determiner (numerically) andat the noun (significant), which is consistent with the resultsfrom the literature (Van Berkum et al., 2005)
21 / 23
Conclusions
I Readers with less cognitive control were more affected bysurprising content (his.fem)
I The results suggest that cognitive control modulates theability to suppress irrelevant predictions (maybe prediction
error?, see: Egner, Monti, & Summerfield, 2010), in this case theability to inhibit the predicted gender form (seinen) when it isnot supported by evidence (seine).
(5) a. predictable
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinenhis.masc
Ofenoven
undand
wartete.waited
b. unpredictable/surprising
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinehis.fem
Mikrowellemicrowave
undand
wartete.waited
22 / 23
Conclusions
I Readers with less cognitive control were more affected bysurprising content (his.fem)
I The results suggest that cognitive control modulates theability to suppress irrelevant predictions (maybe prediction
error?, see: Egner et al., 2010), in this case the ability to inhibitthe predicted gender form (seinen) when it is not supportedby evidence (seine).
(5) a. predictable
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinenhis.masc
Ofenoven
undand
wartete.waited
b. unpredictable/surprising
DerThe
Backerbaker
schobput
denthe
Teigdough
inin
seinehis.fem
Mikrowellemicrowave
undand
wartete.waited
22 / 23
23 / 23