Spacing, Feedback, and Testing Boost Vocabulary Learning in a Web Application

Abstract

Information and communication technology (ICT) becomes more prevalent in education but its general efficacy and that of specific learning applications are not fully established yet. One way to further improve learning applications could be to use insights from fundamental memory research. We here assess whether four established learning principles (spacing, corrective feedback, testing, and multimodality) can be translated into an applied ICT context to facilitate vocabulary learning in a self-developed web application. Effects on the amount of newly learned vocabulary were assessed in a mixed factorial design (3×2×2×2) with the independent variables Spacing (between-subjects; one, two, or four sessions), Feedback (within-subjects; with or without), Testing (within-subjects, 70 or 30% retrieval trials), and Multimodality (within-subjects; unimodal or multimodal). Data from 79 participants revealed significant main effects for Spacing [F(2,76) = 8.51, p = 0.0005, ${\eta }_p^2=0.18$ ] and Feedback [F(1,76) = 21.38, p < 0.0001, ${\eta }_p^2=0.22$ ], and a significant interaction between Feedback and Testing [F(1,76) = 14.12, p = 0.0003, ${\eta }_p^2=0.16$ ]. Optimal Spacing and the presence of corrective Feedback in combination with Testing together boost learning by 29% as compared to non-optimal realizations (massed learning, testing with the lack of corrective feedback). Our findings indicate that established learning principles derived from basic memory research can successfully be implemented in web applications to optimize vocabulary learning.

Keywords: CALL (Computer Assisted Language Learning); distance education; distance learning; language learning; memory; online learning; vocabulary learning; web application.

FIGURE 1

Screenshots of learning and retrieval trials in the web app. Screenshots showing examples of learning trials (A) and retrieval trials (B) as they were displayed to participants in the web applications. Translation for German terms: “Weiter” = “continue.”

FIGURE 2

Screenshots showing feedback screens as they were displayed to participants in the learning application. On the left side is an example of the feedback for correct answers, and on the right side is an example of corrective feedback in case of a wrong response, which also shows the complete word pair and the wrong answer the participant gave. Translation for German terms: “Falsch” = “wrong,” “Richtig” = “correct,” “Die Lösung ist” = “the solution is,” “Ihre Antwort war” = “your answer was,” “Weiter” = “continue.”

FIGURE 3

Estimation plots for the learning principles via Cumming plots. Upper row shows individual participant data in a swarmplot for unpaired data (A, Spacing) and a slopegraph for paired data (B, Feedback, C, Testing, and D Modality). For unpaired data, the mean ± SD are shown as gapped lines. Lower row shows unpaired or paired mean differences as a bootstrap sampling distribution, with the dot indicating the mean difference and the ends of the error bars the 95% confidence interval.

FIGURE 4

Pairwise interaction plots between learning principles. (A–C) Interactions with between-subjects factor Spacing. (D–F) Interactions between the three within-subjects factors Feedback, Testing, and Modality. Values are offset horizontally to avoid over-plotting of error bars. Error bars indicate non-parametrically bootstrapped 95% confidence intervals.

FIGURE 5

Estimation plots and interaction plot for learning and testing direction. Estimation plots for learning direction (A) and testing direction (B). Upper row shows paired individual participant data in slopegraphs. Lower row shows paired mean differences as a bootstrap sampling distribution, with the dot indicating the mean difference and the ends of the error bars indicating the 95% confidence interval. (C) Interaction plot. Values are offset horizontally to avoid over-plotting of error bars. Error bars indicate bootstrapped 95% confidence intervals.