E VIDENCE FOR A SPEED - ACCURACY TRADE - OFF IN BILINGUALS : A DIFFUSION MODEL APPROACH TO THE ATTENTION N ETWORK TASK Joseph M. Burling
Crystal D. Tran
Hanako Yoshida
[email protected]
[email protected]
[email protected]
Observed Response Time Data
◦ There is a lot of ambiguity in the literature on bilingual advantage in cognitive domains such as attention. ◦ Measurements of attention are typically conducted by looking at response times, which are often analyzed by discarding parts of the data and/or interpreting accuracy and time separately. ◦ How do RT and response accuracy interact on the ANT given an individual’s previous language experience, and what is the role of this experience on influencing attention?
Adult, Correct
Model Predicted RT & Accuracy
Adult, Incorrect
3000
0.020 100
Bilingual
Monolingual
0.015 2000
0.010 50
Our Main Contributions
0
0 0.0
0.5
1000
◦ The ANT measures certain aspects of attention such as alerting, orienting, and executive control. ◦ This is tested using specific combinations of cues and congruent/incongruent target orientations. ◦ After viewing a cue, adults used the keyboard to respond to the direction of the target (middle item), while children used a left/right mouse click to respond.
Monolingual & bilingual participants Adult ◦ Bilinguals: ◦ sample size = 31 ◦ age = 24.0 yrs ◦ range = (17, 44) ◦ Monolinguals: ◦ sample size = 32 ◦ age = 25.8 yrs ◦ range = (19, 56)
Child ◦ Bilinguals: ◦ sample size = 57 ◦ age = 79.5 mos ◦ range = (49, 108) ◦ Monolinguals: ◦ sample size = 68 ◦ age = 79.1 mos ◦ range = (45, 117)
0.0
0.5
1.0 Child, Incorrect
1.5
30
0.000 −1.5
−1.0
20
1.5
0.000 10
−5.0 −4.5 −4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0
250
0 0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
RT (sec)
Posterior Distributions of Estimated Diffusion Parameters: Cautiousness, time required to respond
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Child
Adult
Child
Bilingual
Bilingual
Monolingual
Posterior density
9
6
3
Monolingual
60
40
20
0
0 0.8
0.9
1.0
1.1
2.0
2.2
2.4
0.30
2.6
0.32
0.34
0.00
0.02
0.04
0.06
Parameter τ
Parameter α
Initial bias toward correct
Information processing speed
Adult
Child
Adult
12.5
Child
4 Bilingual
Monolingual
Monolingual
Posterior density
10.0
Bilingual
7.5 5.0 2.5 0.0
3
2
1
0 0.7
0.8
0.5
0.6
0.7
0.8
0.9
0
1
2
0.00
0.25
0.50
0.75
Parameter δ
References
Bilinguals:
Monolinguals: 1
α β
τ
Response time |
α, τ, β, δ
Motor response & encoding time
Adult
Posterior density
◦ Parameters were estimated using an extension of the Hamiltonian Monte Carlo algorithm referred to as the No-UTurn Sampler ◦ Five MCMC chains were run in parallel for a total of 10,000 samples, with just as many allowed for initial tuning ◦ Vertical lines display the median of each posterior distribution ◦ In general, bilinguals have estimates which increase RT, except for processing speed δ
β
1.0
Theoretical RT density using estimated model parameters. Inaccurate responses are indicated by a negative RT, while accurate are positive values.
0 0.0
α
0.5
Response time * Accuracy {−1, 1}
δ
π α, τ, δ ∼ µ + tan(unif orm(0, )) 2 β ∼ beta(1.25, 1.25) µ=0
1.0
500
Discussion of Results
Half-cauchy & Beta priors
0.5
0.015
Parameter β
yˆij = wiener(αj , τj , βj , δj ) αj = α0 + α1 ∗ bilingual τj = τ 0 + τ 1 ∗ bilingual βj = β 0 + β 1 ∗ bilingual δj = δ 0 + δ 1 ∗ bilingual
0.0 Child
0.005
0.6
Wiener process likelihood
−0.5
750
The ANT time and accuracy data were modeled as a Wiener/Brownian process in which binary responses (correct/incorrect) and the time in takes to make them are decomposed into cognitive processes such as the speed/accuracy trade-off (α), processing speed of the individual (δ), initial bias toward a particular response (β), and motor response latency (τ ). These parameters determine the shape of the response time distributions for both accurate and inaccurate cases.
Task information & procedure
1.5
0.005
0.010
Bayesian Diffusion Model
Stimuli (adult & child versions)
1.0 Child, Correct
Posterior density
Attention Network Task
Monolingual
Bilingual
1000
◦ Diffusion models allow for simultaneous evaluation of accuracy and response time data, which provides richer information about the underlying processes involved in binary, timed response tasks. ◦ We observed measurable differences in how bilinguals differ from monolinguals when responding to stimuli that measures general attentiveness. ◦ Process models such as these provide additional information about how language influences attention, which can’t be obtained with classical statistical analyses. ◦ Results suggest a higher probability of an incorrect response for biliguals given a speeded response, but delayed RT are overcome by faster processing speed over the monolingual group. ◦ Similar response time statistics between the two groups may be the result of different underlying processes driven by language history.
Adult
0.025
Normalized Density
Introduction to the Problem
0
1
τ
δ
Response time |
1. Vandekerckhove, J., Tuerlinckx, F., & Lee, M. D. (2011). Hierarchical diffusion models for two-choice response times. Psychological Methods, 16, 44–62. 2. Stan Development Team (2014). Stan: A C++ Library for Probability Sampling, Version 2.2. http://mc-stan.org/
0
The above figure illustrates how accurate response times between two groups may look similar (total RT), yet involve different processes in terms of the parameters estimated by the diffusion model. Bilinguals may exhibit slightly slower response times, due to a higher accuracy cost for speeded responses (speed/accuracy trade-off) and lower initial bias; however, RTs may be shortened by faster processing of the information. A history of language switching may generate tendencies for quickly processing information within the context of ANT, yet more evidence about the environment may be required before a bilingual individual can initiate this process.
3. Tran, D. N. & Yoshida, H. (2011). The developmental role of attentional control in language learning. In L. Carlson, C. Holscher, & T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society (p. 1747). Austin, TX: Cognitive Science Society.
Acknowledgments This research was supported in part by University of Houston, a National Institutes of Health grant (R01 HD058620), the Foundation for Child Development (Young Scholars Program), and University of Houston’s Grants to Enhance and Advance Research (GEAR) program. We would also like to thank the children and parents who participated in this study.
Poster Presented at the Boston University Conference on Language Development (BUCLD39)