contrasting facilitation pro les for agreement and re ... · contrasting facilitation pro les for...
TRANSCRIPT
![Page 1: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/1.jpg)
Contrasting facilitation profiles foragreement and reflexives revisited
A large-scale empirical evaluation of thecue-based retrieval model
Lena Jager1, Daniela Mertzen1, Julie Van Dyke2,and Shravan Vasishth1
2University of Potsdam
2Haskins Laboratories
Berlin, September 2018
0
![Page 2: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/2.jpg)
1
IntroductionQuantitative model predictions
ExperimentConclusion
Cue-based retrieval: The ACT-R modelAnderson et al., 2004; Lewis & Vasishth, 2005
Retrieval latency and probability are determined by:
i) Match of the retrieval cues
ii) Similarity-based interference
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 3: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/3.jpg)
2
IntroductionQuantitative model predictions
ExperimentConclusion
Facilitatory interference in ungrammatical sentences
No interference∗The bodybuilder−plur
+ c-com
who worked with the trainer−plur− c-com
injured themselves{plurc-com}.
Interference∗The bodybuilder−plur
+ c-com
who worked with the trainers+plur− c-com
injured themselves{plurc-com}.
RETRIEVAL CUESDISTRACTORTARGET
c-command
plural
plural
c-command
plural
c-commandc-command Facilitation
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 4: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/4.jpg)
2
IntroductionQuantitative model predictions
ExperimentConclusion
Facilitatory interference in ungrammatical sentences
No interference∗The bodybuilder−plur
+ c-com who worked with the trainer−plur− c-com injured themselves{plurc-com}.
Interference∗The bodybuilder−plur
+ c-com who worked with the trainers+plur− c-com injured themselves{plurc-com}.
RETRIEVAL CUESDISTRACTORTARGET
c-command
plural
plural
c-command
plural
c-commandc-command Facilitation
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 5: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/5.jpg)
2
IntroductionQuantitative model predictions
ExperimentConclusion
Facilitatory interference in ungrammatical sentences
No interference∗The bodybuilder−plur
+ c-com who worked with the trainer−plur− c-com injured themselves{plurc-com}.
Interference∗The bodybuilder−plur
+ c-com who worked with the trainers+plur− c-com injured themselves{plurc-com}.
RETRIEVAL CUESDISTRACTORTARGET
c-command
plural
plural
c-command
plural
c-commandc-command Facilitation
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 6: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/6.jpg)
3
IntroductionQuantitative model predictions
ExperimentConclusion
Which cues are used?
→ Implicit assumption of Lewis & Vasishth, 2005:
I All available cues are used equally.
→ No qualitative differences between dependency types.
Dillon et al. (2013). Contrasting intrusion profiles foragreement and anaphora, JML, 69, 85–103.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 7: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/7.jpg)
4
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
I Direct comparison of interference effects in reflexives andsubject-verb agreement.
I Facilitatory interference in subject-verb agreement.
I No facilitatory interference in reflexives.
→ Are structural cues given priority in reflexives?
? Low statistical power.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 8: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/8.jpg)
4
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
I Direct comparison of interference effects in reflexives andsubject-verb agreement.
I Facilitatory interference in subject-verb agreement.
I No facilitatory interference in reflexives.
→ Are structural cues given priority in reflexives?
? Low statistical power.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 9: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/9.jpg)
4
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
I Direct comparison of interference effects in reflexives andsubject-verb agreement.
I Facilitatory interference in subject-verb agreement.
I No facilitatory interference in reflexives.
→ Are structural cues given priority in reflexives?
? Low statistical power.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 10: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/10.jpg)
4
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
I Direct comparison of interference effects in reflexives andsubject-verb agreement.
I Facilitatory interference in subject-verb agreement.
I No facilitatory interference in reflexives.
→ Are structural cues given priority in reflexives?
? Low statistical power.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 11: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/11.jpg)
5
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
Statistical power: 6–30%
I Claim based on a null result in reflexives.I Type M(agnitude) error in agreement conditions?
Dillon et al, 2013 −119 [−205,−33] msMeta-analysis of Jager et al., 2017 −22 [−36,−9] ms
→ see also Vasishth, Mertzen, Jager, & Gelman (2018). Thestatistical significance filter leads to overoptimisticexpectations of replicability, JML.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 12: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/12.jpg)
5
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
Statistical power: 6–30%
I Claim based on a null result in reflexives.
I Type M(agnitude) error in agreement conditions?Dillon et al, 2013 −119 [−205,−33] msMeta-analysis of Jager et al., 2017 −22 [−36,−9] ms
→ see also Vasishth, Mertzen, Jager, & Gelman (2018). Thestatistical significance filter leads to overoptimisticexpectations of replicability, JML.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 13: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/13.jpg)
5
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
Statistical power: 6–30%
I Claim based on a null result in reflexives.I Type M(agnitude) error in agreement conditions?
Dillon et al, 2013 −119 [−205,−33] msMeta-analysis of Jager et al., 2017 −22 [−36,−9] ms
→ see also Vasishth, Mertzen, Jager, & Gelman (2018). Thestatistical significance filter leads to overoptimisticexpectations of replicability, JML.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 14: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/14.jpg)
5
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
Statistical power: 6–30%
I Claim based on a null result in reflexives.I Type M(agnitude) error in agreement conditions?
Dillon et al, 2013 −119 [−205,−33] msMeta-analysis of Jager et al., 2017 −22 [−36,−9] ms
→ see also Vasishth, Mertzen, Jager, & Gelman (2018). Thestatistical significance filter leads to overoptimisticexpectations of replicability, JML.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 15: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/15.jpg)
5
IntroductionQuantitative model predictions
ExperimentConclusion
Dillon, Mishler, Sloggett, & Phillips (2013)
Statistical power: 6–30%
I Claim based on a null result in reflexives.I Type M(agnitude) error in agreement conditions?
Dillon et al, 2013 −119 [−205,−33] msMeta-analysis of Jager et al., 2017 −22 [−36,−9] ms
→ see also Vasishth, Mertzen, Jager, & Gelman (2018). Thestatistical significance filter leads to overoptimisticexpectations of replicability, JML.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 16: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/16.jpg)
6
IntroductionQuantitative model predictions
ExperimentConclusion
Meta-analysis: Interference in ungrammatical conditions
Reflexives
Agreement
−40 −20 0 20 40
Interference effect in ms
Jager, Engelmann, & Vasishth: Similarity-based interference insentence comprehension: Literature review and Bayesian
meta-analysis, JML 94, 2017.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 17: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/17.jpg)
7
IntroductionQuantitative model predictions
ExperimentConclusion
Our study
I Large-sample replication of Dillon et al. (2013)
→ Bayesian parameter estimation.
I Quantitative evaluation of the Lewis & Vasishth (2005)ACT-R cue-based retrieval model.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 18: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/18.jpg)
8
IntroductionQuantitative model predictions
ExperimentConclusion
Model evaluation: the ROPE approach (Kruschke, 2015)
Hyp
oth
etic
ald
ata
A
B
C
D
E
F
-57ms -10ms
1
Model prediction
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 19: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/19.jpg)
9
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R simulations
−400
−100
−20
−5
0
5
20
Inte
rfer
ence
effe
ct (
ms)
I Parameter combinations:I Latency factor F ∈ {0.05, 0.06, ..., 0.6}I Noise parameter ANS ∈ {0.1, 0.2, 0.3}I Maximum associative strength MAS ∈ {1, 2, 3, 4}I Mismatch penalty MP ∈ {0, 1, 2}I Retrieval threshold θ ∈ {−2,−1.5, ..., 0}
I 6000 iterations per parameter configuration
Simulations conducted by Engelmann, Jager, & Vasishth: The effect of prominenceand cue association in retrieval processes: A computational account,https://osf.io/b56qv/
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 20: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/20.jpg)
10
IntroductionQuantitative model predictions
ExperimentConclusion
Ungrammatical conditions from Dillon et al., 2013
Agreement; no interference∗The amateur bodybuilder−plur
+local subj who worked with the personal trainer−plur−local subj
amazingly were{plurlocal subj} competitive for the gold medal.
Agreement; interference∗The amateur bodybuilder−plur
+local subj who worked with the personal trainers+plur−local subj
amazingly were{plurlocal subj} competitive for the gold medal.
Reflexive; no interference∗The amateur bodybuilder−plur
+ c-com who worked with the personal trainer−plur− c-com
amazingly injured themselves{plurc-com} on the lightest weights.
Reflexive; interference∗The amateur bodybuilder−plur
+ c-com who worked with the personal trainers+plur− c-com
amazingly injured themselves{plurc-com} on the lightest weights.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 21: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/21.jpg)
10
IntroductionQuantitative model predictions
ExperimentConclusion
Ungrammatical conditions from Dillon et al., 2013
Agreement; no interference∗The amateur bodybuilder−plur
+local subj who worked with the personal trainer−plur−local subj
amazingly were{plurlocal subj} competitive for the gold medal.
Agreement; interference∗The amateur bodybuilder−plur
+local subj who worked with the personal trainers+plur−local subj
amazingly were{plurlocal subj} competitive for the gold medal.
Reflexive; no interference∗The amateur bodybuilder−plur
+ c-com who worked with the personal trainer−plur− c-com
amazingly injured themselves{plurc-com} on the lightest weights.
Reflexive; interference∗The amateur bodybuilder−plur
+ c-com who worked with the personal trainers+plur− c-com
amazingly injured themselves{plurc-com} on the lightest weights.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 22: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/22.jpg)
11
IntroductionQuantitative model predictions
ExperimentConclusion
Method and Procedure
I Eyetracking-while-reading.
I 181 native speakers of English.
I 48 experimental items from Dillon et al. (2013), Expt. 1.
I Eyelink 1000 (1000Hz) with desktop mount camera.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 23: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/23.jpg)
12
IntroductionQuantitative model predictions
ExperimentConclusion
Bayesian analysis of eye movements
Following Dillon et al., 2013:
I Region of interest: verb/reflexive plus subsequent word
I Dependent variable: total fixation times
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 24: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/24.jpg)
13
IntroductionQuantitative model predictions
ExperimentConclusion
Results
●
●
−80
−40
0
40
Original Replication ACT−R
Inte
rfer
ence
effe
ct (
ms)
● ReflexiveAgreementACT−R
I Similar facilitation profiles inagreement and reflexives.
I Weak support for the Lewis &Vasishth (2005) ACT-R model.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 25: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/25.jpg)
13
IntroductionQuantitative model predictions
ExperimentConclusion
Results
●
●
−80
−40
0
40
Original Replication ACT−R
Inte
rfer
ence
effe
ct (
ms)
● ReflexiveAgreementACT−R
I Similar facilitation profiles inagreement and reflexives.
I Weak support for the Lewis &Vasishth (2005) ACT-R model.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 26: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/26.jpg)
13
IntroductionQuantitative model predictions
ExperimentConclusion
Results
●
●
−80
−40
0
40
Original Replication ACT−R
Inte
rfer
ence
effe
ct (
ms)
● ReflexiveAgreementACT−R
I Similar facilitation profiles inagreement and reflexives.
I Weak support for the Lewis &Vasishth (2005) ACT-R model.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 27: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/27.jpg)
14
IntroductionQuantitative model predictions
ExperimentConclusion
Conclusion
I Very similar estimates for reflexives and agreement.
I Facilitatory interference in both agreement and reflexives ofapprox. 20ms.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 28: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/28.jpg)
15
IntroductionQuantitative model predictions
ExperimentConclusion
Conclusion
I More precise estimates for evaluating the predictions ofquantitative models are needed.
I Larger sample size.I Reduction of measurement error.I Manipulations with larger effects.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 29: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/29.jpg)
15
IntroductionQuantitative model predictions
ExperimentConclusion
Conclusion
I More precise estimates for evaluating the predictions ofquantitative models are needed.
I Larger sample size.
I Reduction of measurement error.I Manipulations with larger effects.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 30: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/30.jpg)
15
IntroductionQuantitative model predictions
ExperimentConclusion
Conclusion
I More precise estimates for evaluating the predictions ofquantitative models are needed.
I Larger sample size.I Reduction of measurement error.
I Manipulations with larger effects.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 31: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/31.jpg)
15
IntroductionQuantitative model predictions
ExperimentConclusion
Conclusion
I More precise estimates for evaluating the predictions ofquantitative models are needed.
I Larger sample size.I Reduction of measurement error.I Manipulations with larger effects.
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 32: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/32.jpg)
16
IntroductionQuantitative model predictions
ExperimentConclusion
Thank you!
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 33: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/33.jpg)
17
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
ACT-R equations
I Retrieval latency of item i : RT := F · e−Ai
I Activation of item i : Ai := Bi + Si + ε
I Baseline activation of item i : Bi := ln(n∑
j=1
t−dj ) + βi
I Spreading activation Si received by item i :
Si :=∑
j∈CuesWjSij)
Sij := MAS − ln(fanj Wj := activation from cue j
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 34: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/34.jpg)
18
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Target Item Distractor Item Retrieval Cues Predictions
+masc
+c-com
masc
c-com
-masc
-c-com
+masc
+c-com -c-com
a.
b.
Inhibitory interference (slowdown) in b vs. abecause the retrieval cue masc matches both items.
TARG
ET-M
ATC
H
-fem
+c-com
-fem
-c-com
-fem
+c-com
+fem
-c-com
c.
d.
Facilitatory interference (speedup) in d vs. c because the retrieval cues fem and c-com match different items.
TARG
ET-M
ISM
ATC
H
masc
c-com
fem
c-com
fem
c-com
Full match
Full match
Partial match
Partial match
No match
No match
Partial match
Partial match
+masc
ambiguous cue
source: Jager, Engelmann & Vasishth, JML, 2015
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 35: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/35.jpg)
19
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
ACT-R prediction: Inhibition in grammatical conditions
Agreement
was { }singular local subjectpersonal trainers - singular
- local subjectbodybuilder + singular + local subject
No interference
personal trainer + singular - local subjectbodybuilder + singular
+ local subject
Interference
was { }singular local subject
cue overload → inhibition
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 36: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/36.jpg)
20
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
ACT-R prediction: Inhibition in grammatical conditions
Reflexives
himself { }singular c-combodybuilder personal trainers - singular
- c-com+ singular + c-com
No interference
bodybuilder himself { }singular c-compersonal trainer + singular
- c-com+ singular + c-com
Interference
cue overload → inhibition
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 37: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/37.jpg)
21
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
ACT-R prediction: Facilitation in ungrammatical conditions
Agreement
bodybuilder personal trainer were{ }plural local subject
- plural - local subject
- plural + local subject
No interference
bodybuilder personal trainers were{ }plural local subject
+ plural - local subject
- plural + local subject
Interference
race → facilitation
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 38: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/38.jpg)
22
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
ACT-R prediction: Facilitation in ungrammatical conditions
Reflexives
bodybuilder themselvespersonal trainer - plural - c-com
- plural + c-com
No interference
bodybuilder themselves { }plural c-compersonal trainers + plural
- c-com- plural + c-com
Interference
{ }plural c-com
race → facilitation
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 39: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/39.jpg)
23
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Bayesian hierarchical regression
Random effects prior distributions:
βsubj , βitem ∼ N4(~0,Cov) (1)
Cov =
σ0
. . .
σ3
· R ·σ0
. . .
σ3
(2)
σ1,...,3 ∼ N+(0, 1) (3)
R ∼ LKJ(2) (4)
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 40: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/40.jpg)
23
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Bayesian hierarchical regression
Random effects prior distributions:
βsubj , βitem ∼ N4(~0,Cov) (1)
Cov =
σ0
. . .
σ3
· R ·σ0
. . .
σ3
(2)
σ1,...,3 ∼ N+(0, 1) (3)
R ∼ LKJ(2) (4)
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 41: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/41.jpg)
24
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Results: Original data
Effect Posterior mean (ms)Dependency 119 [71, 169]Grammaticality 100 [69, 134]Dependency×Grammaticality 9 [-18, 36]
Interference [grammatical] [reflexives] 2 [-57, 60]Interference [grammatical] [agreement] -34 [-85, 15]
Interference [ungrammatical] [reflexives] -18 [-72, 36]Interference [ungrammatical] [agreement] -60 [-112, -5]
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 42: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/42.jpg)
25
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Results: Original data
Effect Posterior mean (ms)
all
Dependency 119 [71, 169]Grammaticality 100 [69, 134]Dependency×Grammaticality 9 [-18, 36]
Mo
del
1 Interference -27 [-56, 1]Dependency×Interference -20 [-46, 6]Grammaticality×Interference -11 [-38, 15]Dependency×Grammaticality×Interference -2 [-27, 24]
Mo
del
2 Interference [grammatical] -16 [-52, 20]Interference [ungrammatical] -38 [-79, 1]Dependency×Interference [grammatical] -17 [-56, 19]Dependency×Interference [ungrammatical] -21 [-56, 12]
Mo
del
3 Interference [grammatical] [reflexives] 2 [-57, 60]Interference [grammatical] [agreement] -34 [-85, 15]Interference [ungrammatical] [reflexives] -18 [-72, 36]Interference [ungrammatical] [agreement] -60 [-112, -5]
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 43: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/43.jpg)
26
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Results: Replication experiment
Effect Posterior mean (ms)Dependency 141 [100, 184]Grammaticality 121 [100, 141]Dependency×Grammaticality -17 [-30, -5]
Interference [grammatical] [reflexives] 12 [-16, 43]Interference [grammatical] [agreement] 5 [-18, 28]
Interference [ungrammatical] [reflexives] -23 [-48, 2]Interference [ungrammatical] [agreement] -22 [-46, 3]
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 44: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/44.jpg)
27
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Results: Replication experiment
Effect Posterior mean (ms)
all
Dependency 141 [100, 184]Grammaticality 121 [100, 141]Dependency×Grammaticality -17 [-30, -5]
Mo
del
1 Interference -7 [-19, 5]Dependency×Interference -2 [-14, 10]Grammaticality×Interference -16 [-30, -2]Dependency×Grammaticality×Interference 2 [-11, 16]
Mo
del
2 Interference [grammatical] 9 [-9, 28]Interference [ungrammatical] -23 [-41, -5]Dependency×Interference [grammatical] -4 [-21, 13]Dependency×Interference [ungrammatical] 1 [-17, 18]
Mo
del
3 Interference [grammatical] [reflexives] 12 [-16, 43]Interference [grammatical] [agreement] 5 [-18, 28]Interference [ungrammatical] [reflexives] -23 [-48, 2]Interference [ungrammatical] [agreement] -22 [-46, 3]
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 45: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/45.jpg)
28
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Total fixation times
Dillon et al., 2013
Large-sample study
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−100 0 100
Interference nested within grammaticality and dependency type
Total fixation times (ms)
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−100 0 100
Interference nested within grammaticality and dependency type
Total fixation times (ms)
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 46: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/46.jpg)
29
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
First-pass reading times
Dillon et al., 2013
Large-sample study
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−40 0 40
Interference nested within grammaticality and dependency type
First−pass times (ms)
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−40 0 40
Interference nested within grammaticality and dependency type
First−pass times (ms)
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives
![Page 47: Contrasting facilitation pro les for agreement and re ... · Contrasting facilitation pro les for agreement and re exives revisited A large-scale empirical evaluation of the cue-based](https://reader034.vdocuments.us/reader034/viewer/2022042211/5eb3149b107ad462d15e8890/html5/thumbnails/47.jpg)
30
IntroductionQuantitative model predictions
ExperimentConclusion
ACT-R EquationsACT-R predictionsBayesian analysisResults
Proportion offirst-pass regressions
Dillon et al., 2013
Large-sample study
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−0.1 0.0 0.1
Interference nested within grammaticality and dependency type
First−pass regressions (propn.)
Int_ungram_agr
Int_ungram_refl
Int_gram_agr
Int_gram_refl
−0.1 0.0 0.1
Interference nested within grammaticality and dependency type
First−pass regressions (propn.)
Jager, Mertzen, Van Dyke, Vasishth Facilitation profiles of agreement and reflexives