An Evidence Accumulation Modeling of Proficiency and Training Effects on Second Language Comprehension in Adults and Adolescents

Abstract

Speaking a second language offers many cognitive benefits¹, but there is a lack ofstudies using computational modeling techniques to study the cognitive mechanisms underlying second language processing. This study uses the Drift Diffusion Model (DDM) to examine whether reading and listening comprehension differs between languages (first versus second), modalities (Picture v. Sound), or proficiency/training. The DDM decomposes cognitive decision-making processes into three main components: Drift Rate(rate of information accumulation), Boundary separation (decision threshold), and Non-decision time (time for encoding and motor response). We tested three hypotheses:

1. Adults: Higher proficiency  would have a higher drift rate compared to  the lower proficiency group in their nonnative language (Ln)
2.  Adolescents: Training would shift the speed-accuracy trade-off towards speed, (lower boundary separation).
3. Both Groups:  Non-decision time would remain consistent in both groups.

The final sample used includes 29 adult participants and 23 adolescent participants. PyDDM was used to fit the DDM, and ANOVAs were performed to statistically analyze the DDM results. Our findings supported our hypotheses:

1. Adult proficiency differences in second language comprehension were attributed to faster and more efficient accumulation of evidence (higher drift rate),
2. Adolescent performance before and after training are attributed primarily to a change in speed accuracy trade off (boundary separation).
3. Non-decision time remained unchanged in both groups

 These findings highlight the distinct processes involved in second language comprehension in adults and adolescents, suggesting that adults benefit from increased automaticity in language processing and adolescents adapt their response strategy from accuracy to speed.

Introduction

As our world becomes increasingly connected and globalized through technology and immigration, bilingualism has become more prevalent². In recent years, the research on bilingualism has moved from measuring the differences between bilinguals and monolinguals³  – following the discovery of the bilingual effect/advantage⁴ and more recent connections between bilingualism and neuroplasticity – to a deeper analysis of second language processing in bilinguals⁵ . Such analysis has led to careful examination of how factors such as proficiency can be linked to contemporary cognitive neuroscience processes like cognitive control^6,7 , or how the modality effect, i.e. changes in learner performance based on the modality (e.g. speech or reading) of the information presented⁸, established in experimental psychology can be applied to native and second language processing through neurolinguistics⁹ .

Previous use of DDM in language sources

While these new studies have been investigating bilingualism and its various effects from the neurolinguistic perspective, little research has applied computational decision-making models to this topic. Proficiency and training play crucial roles in shaping cognitive processes in bilinguals. Higher proficiency is associated with more efficient language processing while training can lead to changes in decision-making processes, both of which can be linked to cognitive mechanisms. Computational decision-making models, namely, the Drift Diffusion model (DDM)  which have been widely used in the field of cognitive science to investigate the decision-making process, provides a framework for examining these effects. In language research specifically, the DDM has proven useful  for investigating certain effects such as semantic relatedness interference¹⁰ , or to analyze individual differences in certain tasks such as differentiating real words from non-real words, making it a relevant framework for examining bilingual language comprehension. The DDM decomposes decision-making performance into the cognitive processes underlying a choice/decision. It uses Reaction Times (RT) and Accuracies to break down decision-making performance into Drift Rate (rate of information gathering), Boundary Separation (speed v. accuracy tradeoff), and Non-Decision time (stimulus encoding and motor response).  This decomposition is useful for studying second language comprehension as it isolates proficiency related differences in information accumulation (Drift Rate) and training related shifts in decision strategies (Boundary Separation). Although the DDM has been applied to find differences between monolingual and bilingual groups¹¹ , it has not been employed yet to uncover individual differences in bilinguals (e.g. as a result of training/proficiency).

This study

This study addresses these individual differences in bilinguals, specifically, the impact of three main factors: language, modality (Picture v. Sound), and proficiency/training on the cognitive processes involved in language comprehension using the DDM. The DDM decomposes reaction times (RT) into constituent cognitive processes: information gathering (Drift Rate), speed vs. accuracy tradeoff (Boundary Separation), and stimulus encoding/motor response (Non-decision Time). To explore these effects, in a semantic animacy judgment task, we started by doing preliminary statistical analysis on the behavioral features (accuracies and reaction times) as a function of each factor (Language, Modality, Proficiency/Training) for two different age groups (Adult and adolescent). Based on the insights gained from the behavioral analyses, we hypothesized that proficiency will be associated with higher drift rates in adults and training will result in lower boundary separation for adolescents. Previous studies have demonstrated that proficiency influences second language comprehension by enhancing lexical access speed¹² enabling faster retrieval of words. This increased speed helps facilitate more rapid accumulation of evidence during comprehension tasks, which aligns with the Drift Rate component of the DDM. We also hypothesized that we wouldn’t find any significant change in Non-decision times because non-decision times reflect mechanical processes such as motor response execution, which are unlikely to vary over the short duration of the study or across consistent participants.  Training in a second language have also shown native-like  electrophysiological signatures in syntactic processing¹³, suggesting that language training can optimize neural mechanisms for efficiency. This aligns with the hypothesis that training can lead to adjustments in cognitive control strategies, which would be reflected in the Boundary Separation component of the DDM. The DDM results were subsequently analyzed using ANOVAs, and a range of visualizations were generated to illustrate key findings.

Results

Behavioral analysis

Adults

A three-way ANOVA was performed on both accuracy and reaction times of the adult dataset with Language and Event Type as within-participant factors and Proficiency as a between-participant factor (Fig 1). The Accuracy ANOVA yielded a significant Language main effect (p=0.000876, L1>Ln), as well as a significant Language x Proficiency interaction (p=0.001). The Reaction Time ANOVA yielded significant main effects in both Event Type (p=0.019, Speech>Reading) and Language (p=0.0000109, Ln>L1) and also a Language x Proficiency interaction (p=0.022).

Due to accuracy being lower than 50%, and thereby lower than chance, Participant 16 (Advanced Proficiency group) was removed from the adult dataset before the DDM Analysis.

Adolescents

A three-way ANOVA was performed on both accuracy and reaction times of the adolescent dataset with Language, Event Type, and Session all as within-participant factors (Fig 2). The Accuracy ANOVA yielded significant Event Type (p=0.00000000235, Reading>Speech) and Language (p=1.1E-18, L1>Ln) main effects, as well as significant interactions: Event Type x Language (p=0.000518) and Language x Session (p=0.025). The Reaction Time ANOVA yielded significant main effects in all three factors: Event Type (p=1.73E-14, Speech>Reading), Language (p=7.01E-12, Ln>L1), and Session (p=0.026, Session 1>Session 2). There were also significant interactions found in Event Type x Language (p=4.97E-11) and Language x Session p=(3.30E-02). There is a visible ceiling effect on accuracy graphs for L1, which reflects the adolescents’ background in familiarity with their native language.

No participants were removed from the Adolescent dataset before DDM Analysis.

DDM analysis

After fitting the DDM Models and obtaining the three parameters of Drift Rate, Boundary Separation, and Non-decision Time, we analyzed the results further through a series of 3-way ANOVAs, on each of the three parameters for both datasets.  Figures and tables were selectively included for significant results to ensure clarity.

Adults

The ANOVA performed on the adult dataset revealed a significant main effect of Proficiency in Drift Rate (p=0.004, Advanced > Basic) and a significant three-way interaction effect of Event Type x Language x Proficiency in the Drift Rate(p=0.044) as well. However, upon plotting the ANOVA as a boxplot (Fig 3), it is immediately apparent that the main effect of Proficiency is much larger than the three-way interaction between all factors in the Drift Rate ANOVA. Thus, these results were followed up with an effect-size test, which yielded a moderate effect size for the main effect of Proficiency (), and a negligible effect size for the three-way interaction (). In summary, the effect of proficiency was apparent only on the drift rate, and there were no significant effects found in Non-decision time and boundary separations.

Adolescents

The ANOVA performed on the adolescent dataset revealed a significant main effect of Event Type ( p = 0.0000000259, Speech>Reading, see Fig 4) and an Event Type x Language interaction (p =  0.000000011, see Fig. 5) in the Drift Rate. However, while statistically significant, these actual effects – as demonstrated by Figures 4 and 5 – are relatively small in magnitude. As a result, our subsequent analysis focused on the significant main effect of Session (p = 0.021, Session1 > Session2, see Fig. 6) in the Boundary Separation. The observed decrease in boundary separation after training (Session 2) suggests that adolescents adopted a faster decision-making strategy after training, in which they prioritized speed over accuracy in the speed-accuracy trade off. This shift may reflect increased familiarity with the task structure. Although training activities were not designed to directly target this tradeoff, there may have been indirect effects such as the training having reduced uncertainties or increased confidence in decision-making, which led participants to require less evidence before committing to a response. There were no significant effects found in Non-decision Time.

Discussion

Overview of study

In this study, we used a Drift Diffusion Model (DDM) to understand the cognitive processes that contribute to language processing in different age groups, sensory modalities, proficiency levels, and languages (first versus second). We were interested in determining what (if any) cognitive processes were behind differences in decision-making performance, and whether they differed between age groups. We were also interested in seeing if these effects were dependent on modality, language, proficiency, or training.

Our results revealed that different cognitive processes underpinned performance differences in the two groups. In adults, the higher drift rate observed in the Advanced group compared to the Basic group in their non-native language (Ln) confirms our primary hypothesis. For adolescents, the lower boundary separation after training supports our hypothesis regarding changes in speed-accuracy trade-off. Lastly, consistent with our secondary hypothesis, we found that non-decision time had no significant changes across both groups and all conditions suggesting that stimulus encoding is minimally affected by language, proficiency, or modality.

This study makes several important contributions to the fields of second language processing and cognitive modeling. Firstly, by applying the DDM to language processing, we are providing a quantitative framework for understanding how cognitive processes like evidence accumulation (Drift Rate), speed-accuracy trade-off (Boundary Separation) and Non-decision time influence decision-making in second language processing in bilingual individuals. Secondly, our findings of higher drift rate in higher proficiency adults are consistent with theories of automaticity which can provide support for the idea that higher proficiency can reduce cognitive load to improve processing speed. Lastly, for adolescents, the shift towards lower boundary separation after training shows how decision-making strategies can adapt over time in the context of second language processing which contributes to the concept of plasticity of cognitive processes in language learning.  These contributions have important implications for evaluating the effectiveness of language learning programs and strategies.

Adults Behavioral

Behavioral performance from the adult dataset showed that there was a Language main effect for both the reaction time (Ln>L1) and accuracy (L1>Ln), which makes sense given that Ln is a nonnative language and therefore had longer reaction times and lower accuracies¹⁴ . They were also slower in responding to Speech events in comparison to Reading events, suggesting that the differential encoding processes for speech and text may contribute to the observed effect – also known as the modality effect⁸ . In addition, there were more trials for pictures in comparison to sound (around two times more), this task design with extra practice in picture tasks may have created a sort of practice effect, where higher repetitions of picture tasks primed participants to answer faster¹⁵ . However, this result was not preserved in the DDM, so this effect and the extent to which experimental conditions contributed to it remains to be tested in future work.

Beyond the main effects, the Language x Proficiency interaction was also found to be significant in both reaction time and accuracy. While both groups exhibited longer reaction times for the non-native language (Ln) in comparison to the native language (L1) they differed in patterns of performance. The basic proficiency group had faster reaction times for L1 (basic RT< advanced RT) in comparison to the advanced proficiency group, while the converse was true for Ln (advanced RT< basic RT). This interaction effect demonstrates the more efficient processing of Ln stimuli by more advanced participants in comparison to the basic participants. This pattern suggests that as language proficiency increases, the processing demands for Ln decrease, which is consistent with theories in second language acquisition where higher proficiency is associated with reduced processing costs¹⁶ .

Adults DDM

The Adult DDM ANOVA found a main effect of Proficiency in Drift Rate where advanced participants had higher drift rates than basic participants. This is consistent with the drift rate representing the relative amount of available information. In the context of language processing, drift rate can be understood as the speed and efficiency of lexical information accumulation during decision-making tasks¹⁷.  .

These higher drift rates could be interpreted through the lens of automaticity in second language processing. This framework posits that as proficiency in language increases, the processing speed of linguistic information also increases and less cognitive effort is generally needed for this processing. The underlying reason for this enhanced processing ties back to the increased amount of information available on the language, which allows for more connections between words and faster retrieval of relevant information¹² . However it should also be noted  that the moderate effect size for drift rate suggests that while proficiency does play a role in affecting drift rate there may be other factors that impact drift rate as well.

The absence of boundary effects and Non-decision time effects indicates that there was no significant change in speed-accuracy tradeoff or encoding/motor response execution within/between any groups. This suggests that differences in performance can be primarily attributed to the increased amount of information available, and efficiency in information processing. This pattern of results, specifically the main effect of Proficiency on Drift Rate can be viewed as a quantitative measure of the concept of automaticity in second language processing.

Adolescent Behavioral

The adolescent dataset exhibited a more complex pattern of results, with multiple significant main and interaction effects. Consistent with the adult dataset, main effects of language (Ln > L1) and Event Type (Speech > Reading) were observed in reaction time. The ceiling effect in accuracy for the native language (L1) corroborates the participants’ background as native speakers receiving formal education in this language¹⁸ .

Beyond this, the main effect of Language (Ln>L1) and Event Type (Speech>Reading) in Reaction time is also consistent with the adult dataset results, which could be a result of either the same discrepancy in the number of Speech and Reading trials or differential encoding processes. The interaction effects Event Type x Language and Language x Session were found in both Reaction time and Accuracy, which suggests that Language plays a relatively large role in the interaction between factors.

Most prominently, Session/Training only had an effect on the reaction time, and accuracy did not improve as a result of a second session. This effect was preserved and explained through the DDM as discussed below.

Adolescent DDM

Adolescents generally had lower boundary separation in Session 2 compared to Session 1. This suggests that in the second session, participants generally chose to prioritize speed over accuracy in their speed-accuracy tradeoff. Although there were only two sessions, this change in speed-accuracy tradeoff may be due in part to the participant’s knowledge on what the second session would be testing on from their experience in the first session. This could reflect a sort of “practice” or task familiarity effect where the first trial questions primed the adolescent group for the second trial performance¹⁹ . This pattern of results indicates that there was a change in cognitive decision making processes where adolescents made a strategic shift to speed over accuracy rather than an increase in information like the adult group had.

In both Adult and Adolescent groups, there were also no significant changes in Non-decision time, as previously hypothesized. This was expected because the factors tested within this study have minimal relationship with the time dedicated to stimulus encoding which generally remains constant under consistent conditions. However, the observed differences in reaction times between modalities (Speech and Reading tasks) could have been attributed to non-decision time effects. The absence of non-decision time differences suggests that, even though the encoding processes for auditory and visual information are distinct, their durations do not differ substantially enough to impact the time allocated to Non-decision components, such as stimulus encoding or motor preparation.

Limitations and future work

Our study had some limitations that could be addressed in future research. A notable limitation was the imbalance of the number of trials between modalities, with picture trials being twice as frequent as sound trials. This disparity, as mentioned previously in the behavioral analysis, may have biased our results towards visual processing. Individual differences in cognitive control, attention and working memory may have also influenced results. Future studies could include measures of these cognitive abilities as a supplement to decision-making models to better understand their role in language processing. Additionally, although this study included different age groups, there were different procedures for each age group, which limited our ability to compare the two groups. By performing the same procedure on different age groups, we would be able to study the effects of age on language processing as another factor in this study. The experimental design could be further refined to balance trials across modalities to help isolate modality specific effects. Future studies could also explore the role of other cognitive models such as the Linear Ballistic Accumulator model (LBA).

The generalizability of our findings is limited by the relatively small sample size and the specific demographic of our participants. While this group can provide valuable insights into the effects of language training on decision-making these findings may not extend to other populations with different linguistic or cultural backgrounds, proficiency levels or educational contexts. Future studies should aim to include more diverse samples to enhance validity of results. A study with languages from a different language family from both Basque and Spanish could help provide more insights into these effects.

Methods

Data source

The data used in this study was obtained through direct correspondence with the original researchers (Kshipra Gurunandan, Jaione Arnaez-Telleria, Manuel Carreiras, and Pedro M. Paz-Alonso) whose work was published in the Journal of Neuroscience in 2020²⁰ . The rest of the dataset is also publicly available on the OpenNeuro database(openneuro.org). The original study collected both behavioral and fMRI data from two groups (adults and adolescents) and two tasks (language comprehension and language production) with aim to explore differential specialization and plasticity in language systems.

For both experiments, language proficiency was assessed using a modified version of the Boston Naming Test, controlled for cognates across Spanish, Basque, and English.

This study will focus only on the language comprehension task using behavioral data from the two groups.

Language Comprehension Task

The language comprehension task consisted of two parts, reading comprehension and speech comprehension. The language comprehension task was a semantic animacy judgment task, in which participants were instructed to indicate as quickly and accurately as possible whether the stimuli presented were living or nonliving by pressing buttons using their dominant hand. Stimuli were presented either visually through a single-word text (reading comprehension) on a black screen or auditorily through headphones (speech comprehension). Participants completed this task in both their native language (Spanish, L1), and their non-native language (Basque, Ln).

Datasets

Participants who had insufficient behavioral data were removed during behavioral data analysis. Participants who had less than 50% accuracy were removed before the DDM was implemented to ensure data reliability, though we acknowledge that in small datasets, removing participants can influence findings. Despite this limitation, the total observations remains above 8000, ensuring statistical power for model estimation. As a result, from 33 original Adult participants, 4 Adult Participants were removed for 29 final Adult Participants. No Participants were omitted in the Adolescent dataset due to sufficient behavioral data and accuracy, making for 23 total Adolescent Participants.

The final Adult dataset comprises 29 native Spanish-speaking adults enrolled in Basque language courses at the same institution, with 15 participants at the advanced level and 14 at the basic level.

The final Adolescent dataset comprises 23 native Spanish adolescents, who all take part in a 3-month-long after-school English immersion program. These participants are all native Spanish speakers from Basque Country in Spain acquiring Basque in school and studying at schools where they were taught in both Spanish and Basque, and also English as a foreign language.

The key difference between the datasets is that while the Adults were split into two groups based on Proficiency (Advanced vs. Basic), the Adolescents were all tested before and then after a period of 3 months and therefore are categorized based on Session (Session 1 and Session 2). These methodological differences between the two age groups could introduce variability in comparison, limiting the ability to compare across age groups.

The final datasets both consist of over 8000 observations, across 3 Experimental Conditions and 2 Dependent behavioral measurements:

Experimental Conditions

• Event Type: Indicates whether task was reading comprehension or speech comprehension
• Language: Indicates whether task was in native Spanish (L1) or non-native Basque (Ln) language
• Proficiency (Adults dataset only): Indicates which group (Advanced or Basic) each participant belongs to
• Session (Adolescents dataset only): Indicates if the observation was taken before training (Session 1) or after 3 months of training (Session 2)

Dependent Behavioral measurements

• Accuracy: Indicates whether Participant’s response was correct (1) or incorrect (0)
• Reaction Time: Indicates reaction time in seconds accurate to four decimal places.

For each participant, there were approximately 92-96 trials per “picture” Event Type, and approximately 45-48 trials per “sound” Event Type, for both L1 and Ln languages. This imbalance presents a limitation, which, while inherent to the dataset, could introduce confounding variables such as practice effects or modality-based biases.

DDM

The Drift Diffusion Model

The Drift Diffusion Model (DDM) is a computational model that analyzes reaction times and accuracies or choices to study the underlying cognitive mechanisms of decision making between two options²¹ . First described 50 years ago, it is now widely used in the field of cognitive neuroscience. The DDM is a part of a larger class of Evidence accumulation models, which conceive decision making as a process in which information, “evidence”, is accumulated over time in favor of one option or another, until enough evidence is accumulated to reach a threshold in favor of a specific response²² .

The standard DDM includes four key components: starting point (z), drift rate (v), boundary separation (a) and Non-decision time (Ter). The starting point represents response bias. For example, if an individual is more biased to response A in comparison to response B, their starting point will reflect this by starting closer to the response A threshold than the response B threshold. The drift rate is the rate of evidence accumulation. A higher (lower) drift rate represents a greater (lower) rate of evidence accumulation and visually, a “steeper (lesser) slope” of evidence. The boundary separation represents the distance between the response thresholds. This value models the speed-accuracy trade-off: smaller (larger) boundary separation means that there is less (more) evidence needed to reach a threshold and is, therefore, a less (more) accurate decision made in a faster (slower) time. The Non-decision time is time that is independent of the decision-making process, and instead represents the time that the participants take to encode the stimulus presented and to execute motor responses such as pressing a button to indicate a response.

In the context of this study, the DDM is suited for understanding cognitive processes because it models the decision-making process using behavioral data – such as reaction time or accuracy – that has already been collected but has not been analyzed in this context yet. However, it should be noted that the DDM assumes decision-making is captured by these drift rate, boundary separation and non-decision time, which may not fully account for other factors of language comprehension such as working memory or cognitive load.

Implementation

The particular form of DDM used in this study was the accuracy DDM, where the starting point is 0 because participants are not biased towards a correct or incorrect option. The other components were set as ‘free parameters’, i.e. initially varying between certain intervals and ultimately fitted by the DDM to the behavioral data (accuracies and RTs). The particular intervals for the parameters used were: Drift: (-2,2), Boundary Separation: (0.2,2), and Non-decision Time: (0.2, 2). These parameter ranges were found using PyDDM’s interactive GUI to manually explore how well different parameter ranges fit to the behavioral data, by adjusting the sliders for drift rate, boundary separation, and non-decision time.  We ran three implementations of the DDM and chose the one with the best fit but results were qualitatively very similar across implementations. Four separate DDMs were then performed on the data of each participant, for each of the combinations of Event Type x Language.

The model used in this study was implemented using the Python package PyDDM, first referenced in a paper by Maxwell Shinn, Norman H Lam, and John D Murray²³ . Within PyDDM, the model used was the generalized DDM, referred to as GDDM in the package.

Statistical Analyses

 The behavioral data was first analyzed on a participant level by taking the median reaction time and average accuracy of each participant for each combination of Event Type and Language (ex: Sound-L1, Sound-Ln etc.). The reaction time and accuracies were then plotted on separate graphs for each of the datasets. The median was used for reaction times instead of average (like accuracies were) because reaction times are skewed to the right. Effects and interactions were considered significant when p-values were less than 0.05.

The three assumptions of ANOVA, normality, homogeneity of variance and independence were verified before the ANOVAs were performed.  A series of three-way ANOVAs were then performed on the behavioral data, using the previously found median reaction times and average accuracies for each condition.  Outlying RTs were first identified and then removed before the analysis.

A series of three-way ANOVAs were also performed on the DDM outputs. Box Plots, Interaction plots, and dot plots were used to visualize statistically significant results. The adult dataset, Event Type (Sound, Picture) and Language (L1, Ln) were within-participant factors, while Proficiency (Advanced, Basic) was a between-participant factor. For adolescent dataset, Event Type (Sound, Picture), Language (L1, Ln) and Session (1, 2) were all within-participant factors.

Separate two-way and one-way ANOVAs were then performed on significant interactions found in the three-way ANOVA. Effect sizes () for statistically significant interaction were also found to provide more context on the practical significance of findings.

Python was used for compiling, analyzing, and visualizing behavioral data. The pandas package was used for compiling and reorganizing data, and seaborn and matplotlib.pyplot were used for visualization. Simple two-way and one-way ANOVAs were conducted with Python package pingouin. Preprocessing, Reshaping, and Three-way ANOVAs were performed in R using rstatix, tidyr and dplyr libraries. Their plots were created using the ggplot2 library, and effect sizes were calculated using the effectsizes library.

References

1. V. Marian, A. Shook. The Cognitive Benefits of Being Bilingual. Cerebrum.13 (2012). [↩]
2. S. Dietrich, E. Hernandez. Language Use in the United States: 2019. Census.gov. Available from: https://www.census.gov/library/publications/2022/acs/acs-50.html (2022). [↩]
3. E. Bialystok. Bilingualism. Wiley Interdisciplinary Reviews: Cognitive Science. 1, 559–72 (2010). [↩]
4. E. Peal, W. E. Lambert. The Relation of Bilingualism to Intelligence. Psychological Monographs: General and Applied 76, 1–23 (1962). [↩]
5. J. F. Kroll, E. Bialystok. Understanding the consequences of bilingualism for language processing and cognition. Journal of Cognitive Psychology. 25, 497–514 (2013). [↩]
6. S. Kheder, E. Kaan. Cognitive control in bilinguals: Proficiency and code-switching both matter. Cognition. 209, 104575 (2021). [↩]
7. A. Luque, K. Morgan-Short. The relationship between cognitive control and second language proficiency. Journal of Neurolinguistics. 57, 100956 (2021). [↩]
8. P. Ginns. Meta-analysis of the modality effect. Learning and Instruction. 15, 313–31 (2005). [↩] [↩]
9. D. C. Gallagher, M. Yano, S. Ohta. The Neurophysiological Modality Effect in Native and Second Language Processing: An ERP Study. bioRxiv (Cold Spring Harbor Laboratory). (2022). [↩]
10. L. Todorova, D. A. Neville, V. Piai. Lexical-semantic and executive deficits revealed by computational modelling: A drift diffusion model perspective. Neuropsychologia. 146, 107560 (2020). [↩]
11. G. Ong, D. K. Sewell, B. Weekes, M. McKague, J. Abutalebi. A diffusion model approach to analysing the bilingual advantage for the Flanker task: The role of attentional control processes. Journal of Neurolinguistics. 43, 28–38 (2017). [↩]
12. N. Segalowitz, J. Hulstijn. Automaticity in Bilingualism and Second Language Learning. In: Handbook of bilingualism: Psycholinguistic approaches. (2005). [↩] [↩]
13. K. Morgan-Short, K. Steinhauer, C. Sanz, M. T. Ullman. Explicit and Implicit Second Language Training Differentially Affect the Achievement of Native-like Brain Activation Patterns. Journal of Cognitive Neuroscience 24, 933–947 (2012). [↩]
14. O. Garcia, N. Faghihi, A. R. Raola, J. Vaid. Factors influencing bilinguals’ speed and accuracy of number judgments across languages: A meta-analytic review. Journal of Memory and Language. 118, 104211 (2021). [↩]
15. T. E. Goldberg, P. D. Harvey, K. A. Wesnes, P. J. Snyder, L. S. Schneider. Practice effects due to serial cognitive assessment: Implications for preclinical Alzheimer’s disease randomized controlled trials. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring. 1, 103–11  (2015). [↩]
16. P. Wang, X. Huang, X. Chang. The effect of inhibitory control and language proficiency on intra-sentential switching costs in reading comprehension. Acta Psychologica. 241, 104063–3 (2023). [↩]
17. R. Ratcliff, G. McKoon. The Diffusion Decision Model: Theory and Data for Two-Choice Decision Tasks. Neural Computation. 20, 873–922 (2008). [↩]
18. O. Garin. Ceiling Effect. Encyclopedia of Quality of Life and Well-Being Research. (2014). [↩]
19. J. Scharfen, D. Blum, H. Holling. Response Time Reduction Due to Retesting in Mental Speed Tests: A Meta-Analysis. Journal of Intelligence. 6, 6 (2018). [↩]
20. K. Gurunandan, J. Arnaez-Telleria, M. Carreiras, P. M. Paz-Alonso. Converging Evidence for Differential Specialization and Plasticity of Language Systems. The Journal of Neuroscience. 40, 9715–24 (2020). [↩]
21. R. Ratcliff. A theory of memory retrieval. Psychological Review. 85, 59–108 (1978). [↩]
22. C. E. Myers, A. Interian, A. A. Moustafa. A practical introduction to using the drift diffusion model of decision-making in cognitive psychology, neuroscience, and health sciences. Frontiers in Psychology. 13, 1039172 (2022). [↩]
23. M. Shinn, N. H. Lam, J. D. Murray. A flexible framework for simulating and fitting generalized drift-diffusion models. eLife. 9, e56938 (2020). [↩]