Abstract
This article presents the design and a first pilot evaluation of the computer-based training program Calcularis for children with developmental dyscalculia (DD) or difficulties in learning mathematics. The program has been designed according to insights on the typical and atypical development of mathematical abilities. The learning process is supported through multimodal cues, which encode different properties of numbers. To offer optimal learning conditions, a user model completes the program and allows flexible adaptation to a child's individual learning and knowledge profile. Thirty-two children with difficulties in learning mathematics completed the 6-12-weeks computer training. The children played the game for 20 min per day for 5 days a week. The training effects were evaluated using neuropsychological tests. Generally, children benefited significantly from the training regarding number representation and arithmetic operations. Furthermore, children liked to play with the program and reported that the training improved their mathematical abilities.
Introduction
Developmental Dyscalculia (DD) is a persistent difficulty in learning or comprehending arithmetic, affecting a significant portion of the school-age population. While many interventions exist, they often suffer from two primary flaws: they are not grounded in the hierarchical "four-step" model of numerical development, and they lack the granularity to adapt to a child's specific "knowledge gaps."
The authors developed Calcularis, a neuro-cognitively grounded computer training program designed to bridge the gap between basic number sense (like estimating quantities) and complex arithmetic (like bridging through ten). By using a sophisticated user model, the program treats learning as a non-linear journey through a "skill net," providing a highly personalized learning path for children with dyscalculia.
The Calcularis Framework
The design of Calcularis stands on three core pillars:
- Neuro-cognitive Scaffolding: The program follows the "triple-code model," transitioning children from concrete sets of objects to verbal/Arabic symbols, and finally to a spatial mental number line. This hierarchical approach ensures that foundational skills are consolidated before moving to more complex operations.
- The Student Model (Dynamic Bayesian Network): Unlike static software, Calcularis models a child’s knowledge across 100 distinct mathematical skills. It uses a Dynamic Bayes Net to estimate the mastery probability for each skill based on performance, deciding whether to "Stay," "Go back" to a precursor skill, or "Go forward."
- Multimodal Cues and the Bug Library: The program uses consistent color-coding (units=green, tens=blue, hundreds=red) to provide a visual "anchor" for abstract digits, facilitating the acquisition of the Arabic place-value system. Additionally, the system identifies specific error patterns (e.g., "inverting digits" like writing 52 for twenty-five) and triggers surgical remediation tasks.
Methods and Evaluation
The pilot study involved 41 children (grades 2–5) with math difficulties, split into a training group and a waiting-list control group. The training group used Calcularis for 20 minutes per day, 5 days a week, for 6 to 12 weeks.
Evaluation utilized several standardized and custom benchmarks:
- Heidelberger Rechentest (HRT): A standardized German test for arithmetic speed and accuracy.
- Computer-based Tests: Custom assessments for non-symbolic magnitude comparison, number line positioning (0–10, 0–100, 0–1000), and arithmetic operations.
- Case Study Data: Detailed logs of individual training trajectories to demonstrate how the adaptive algorithm handles different learning speeds.
Key Results and Insights
The study yielded several significant findings regarding the efficacy of the training:
- Significant Arithmetic Gains: The training group showed a significant improvement in subtraction with a medium-to-large effect size ($r = 0.39$ in computer tests; $r = 0.52$ in the HRT). Interestingly, subtraction was found to be a much stronger indicator of a child's spatial number line representation than addition.
- Number Line Refinement: Children became significantly more accurate at placing numbers on a line. In the 0–10 range, the deviation from the correct position was reduced by 33% after just 6 weeks.
- Dosage and Engagement: While 6 weeks showed clear benefits, extending the training to 12 weeks yielded additional trends of improvement, particularly in addition skills. Furthermore, children reported high levels of engagement, averaging 3.0 out of 4.0 on a satisfaction scale.
Conclusion
The Calcularis pilot study demonstrates that a neuro-cognitively grounded, adaptive training program can significantly enhance numerical cognition in children with math difficulties. While the study faced limitations such as ceiling effects in certain tasks and the use of a waiting-list control, the results are promising. Future work will focus on integrating more explicit motivational features and refining the "bug library" based on large-scale data to further optimize individualized learning trajectories.
Summary updated