Benjamin GaglJul 21, 2023
🚨🚨NEW PREPRINT 🚨🚨 "Investigating lexical categorization in visual word recognition based on a joint diagnostic and training approach for language learners." https://psyarxiv.com/rs6gy/
We designed a training program including a machine learning-based diagnostic approach to train lexical categorization optimally, i.e., the process most likely implemented in the visual word form area (https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1009995). 1/N
Show threadBenjamin GaglJul 21, 2023
We trained language learners of German from various language backgrounds with a lexical decision task, including feedback.
2/N
Show threadBenjamin GaglJul 21, 2023
We found significant training effects on reading skills across three studies, two of which were RTCs, and we preregistered the final study.
3/N
Show threadBenjamin Gagl
In addition, we developed a diagnostic procedure intending to detect the responders to the training based on explainable machine learning. Here we investigated various variants (i.e., changing feature creation procedures or applying different ML algorithms) to detect the optimal combination—most investigated procedures produced a medium-to-high correlation of predicted vs. observed training effects.
4/N
Jul 21, 2023 at 10:08amWeb
Show threadBenjamin GaglJul 21, 2023
Interestingly, when inspecting the results of the feature selection process, features based on the central model parameter of the Lexical Categorization Model (LCM) have been the essential word-level features for the predictive models.
5/N
Show threadBenjamin GaglJul 21, 2023
The best predicting model resulted in a correlation of .69 predicted vs. observed reading skill increase. When applied, the training effect on the group level increased from 23% to 43%.
6/N
Show threadBenjamin GaglJul 21, 2023
We concluded that the exact description of neuro-cognitive processes in relevant brain regions can motivate the development of new and effective training methods and aid machine learning-based diagnostic procedures for detecting training responders.
7/N