An individualized machine-learning diagnostic using LCM-based features predicts lexical categorization training benefits with moderate-to-good accuracy¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2024 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Aggregating predictions across all machine-learning pipelines, the correlation between predicted and observed training effect was moderate, r = 0.58. → Benjamin Gagl & Klara Gregorová 2024 q3 i? The best-performing pipeline reached a predicted–observed correlation of 0.69, explaining nearly 50% of training-effect variance (R² = 0.476). → Benjamin Gagl & Klara Gregorová 2024 (2)

Evidence¶

Benjamin Gagl & Klara Gregorová 2024¶

Benjamin Gagl & Klara Gregorová. (2024). Investigating lexical categorization in reading based on joint diagnostic and training approaches for language learners. npj Science of Learning 9:29. https://doi.org/10.1038/s41539-024-00237-7

q3 · i? · causal · r2

Prediction models were built with consensus nested leave-one-out cross-validation using pre-test reading speed and first-session data only; individual pipeline correlations ranged from −0.10 to 0.69 with a mean of 0.42. The aggregated prediction achieved r = 0.58.

"When we aggre- gate the predicted values from all pipelines for each participant (median across 720 predictions), the correlation between pre- dicted vs. observed training effect was moderate ( r = 0.58; t(73) = 6.02, confidence interval: 0.40–0.71, p < 001)."

Benjamin Gagl & Klara Gregorová 2024 (2)¶

Benjamin Gagl & Klara Gregorová. (2024). Investigating lexical categorization in reading based on joint diagnostic and training approaches for language learners. npj Science of Learning 9:29. https://doi.org/10.1038/s41539-024-00237-7

q3 · i? · causal · r2

The best pipeline combined multiple regression fitting, a feature-selection cutoff of 10, and a predictor composition estimating change of lexical categorization uncertainty and OLD20 effects within the first session; it was best in 21 of 75 cross-validation loops, and its predictor composition was best in 55 of 75 loops.

"the final correlation of pre- dicted vs. observed lexical categorization training effects resulted in 0.69 (t(73) = 8.16, con fidence interval = [0.55–0.79], p < 0.001, see Fig. 5A), explaining nearly 50% of the variance of the training effect (R2 = 0.476)."

Discussion¶

Related Claims¶

• Lexical categorization training significantly increases reading speed in adult non-native German learners, replicated across three experiments — related
• LCM-related features, especially lexical categorization uncertainty, are the most important word-characteristic predictors of training success — related
• Training response times rise with lexical categorization uncertainty and fall across sessions, but a reduction of the uncertainty effect across sessions appears only in Experiment 1 — related
• Machine-learning-based responder selection raises mean group reading-speed gain from 23% to 43% — related