Overview of Osgood’s predictions for proactive memory effects, variable coverage in this study, conditions, stimuli, counterbalance procedures, and experiments.

(a) We depict how our design would look if it confirmed Osgood’s (1949) proposed surface. Cue and target relatedness are shown along the y- and x-axes, respectively, while memory change is shown along the z-axis, relative to the Control condition, which is simply depicted as the z = 0 surface. (b) In the narrower stimulus set, variables covered the full range of associative strength (AS) values. Plus signs along the ΔCue (green) and ΔTarget (blue) lines depict how cue and target relatedness were respectively distributed. Purple plus signs inside the scatterplot show how cue and target relatedness values were distributed in the ΔBoth condition. (c) Pairs were divvied into four experimental conditions during supplemental pair learning (which was performed first), followed by base pair learning, which included a fifth Control condition. After 5-mins or 48-hrs, base and supplemental pairs were tested (in that order). (d) Each base pair was included in a different condition for every five subjects. (e) The primary four experiments involved crossing delays and stimulus sets. In (a) and (b), example word pairs from (c) are labeled for explanatory purposes. See also Fig 1-Supp 1 for visualizations using the stimulus set with wider semantic relatedness.

Stimulus set, test delay, and experimental condition all determined whether PF or PI occurred.

The narrow stimulus set included only single-step semantic relations between base and supplemental cues and targets (top row), whereas the wide stimulus set included the full range of semantic relationships found in the English language (bottom row). Delays were either 5 min (left column) or 48 hrs (right column). All comparisons were significant except those labeled with connected gray bars and either ‘ns’ (p > 0.1) or † (0.05 < p < 0.1), and sets of three connected bars indicated all ‘ns’ differences. Data points from individual subjects were jittered for improved visualization. See Fig 2-Supp 1 for supplemental pair memory.

Memory and interdependence in the ΔTarget condition as a function of relatedness.

(a) We plotted each pair’s across-subject memorability value against its corresponding target-Δtarget semantic relatedness in all experiments. AS and GloVe values were used in the top and bottom rows, respectively. In the wider relatedness stimulus set, 48-hr experiment, semantic relatedness proactively improved memory. Furthermore, PI occurred under conditions of low relatedness and PF under high relatedness. (b) We plotted memory dependence for each base pair-supplemental pair target duo against its relatedness in all experiments. High dependence means that subjects tended to remember or forget both targets in the duo (relative to recalling one and not the other). Relatedness significantly increased dependence in both 48-hr experiments (p < 0.05) but this was marginally significantly in the 5-min experiments (0.05 < p < 0.1). Thick dotted lines show the threshold of dependence levels against all other pairs at the 95th percentile. See also Fig 3-Supp 1 for intrusion data from this condition.

Semantic relatedness between cues benefits memory in the 48-hr experiments and boosts interdependence in all experiments.

(a) Across-subject memorability for each pair in the ΔCue minus Control condition benefited from cue relatedness in the 48-hr, but not 5-min, experiments [top: AS; bottom: GloVe]. (b) Relatedness increased memory interdependence in all experiments. Thick dotted lines show the threshold of dependence levels against all other pairs at the 95th percentile.

Increasing cue + target relatedness boosted interdependence but not memorability in the ΔBoth condition across experiments.

(a) We plotted cue + target relatedness against memorability in each experiment. Memorability did not increase with relatedness in any experiment. (b) Memory interdependence increased with cue + target relatedness in all experiments except the wider stimulus set, 5-min delay experiment. See Fig 5-Supp 1 for smoothed, 2-D plots that plot cue and target relatedness as separate variables predicting memorability and interdependence.

Osgood-style surfaces for proactive effects and interdependence.

(a) We plotted memorability from all conditions minus the Control condition from all experiments in 3-D coordinates. These plots had cue and target relatedness on the y- and x-axes, respectively. On the z-axis, we plotted base pair memorability, oriented as PF for positive change and PI for negative change. (a, right) For the narrower stimulus set, 48-hour delay experiment, memorability for the No Δ (± across-pair standard deviation) is at the cue identity, target identity corner point (red circle). We plotted ΔTarget condition memorability along the cue identity line against target relatedness (blue) and ΔCue condition memorability along the target identity line against cue relatedness (green) (both using ± standard error from the ordinary-least-squares regression fit). We plotted ΔBoth condition memorability as a locally smoothed surface as a function of both cue and target relatedness (purple). Transparent surface grids above and below zero represent p < 0.01 significance boundaries from permutation tests; significant points on the surface are indicated by a darker shade of purple. (left) We created similar plots for all other experiments. (b) Interdependence measurements for all experiments and conditions formatted similarly to (a).

Base pair learning differed based on stimulus set and condition and generally benefited from higher semantic relatedness.

(a) In the narrower stimulus set, learning time (mean trials to criterion) followed this trend: No Δ < ΔCue < ΔTarget < ΔBoth < Control. The wider stimulus set tended to show nearly the same pattern: No Δ < ΔCue </= ΔTarget < ΔBoth = Control, where ‘</=’ indicates a significant difference in one experiment but not the other. All comparisons were significant except those labeled with gray bars and ‘ns’ (p > 0.1). Data points from individual subjects were jittered slightly for visualization, and a few outliers (slower learners) were omitted from the plot, for better visualization. (b-d) We correlated relatedness with learning times [top: AS; bottom: GloVe]. (b-c) Learning time across subjects for each word pair generally decreased with increasing cue relatedness in the ΔCue condition (b) and decreased with increasing target relatedness in the ΔTarget condition (c). (d) In the ΔBoth condition, learning time generally decreased with cue + target relatedness in the narrower stimulus set, but not in the wider stimulus set. In (b-d), Pearson correlations are shown in the plots followed by * when p < 0.05 and ** when p < 0.01.

Under otherwise identical conditions, memory benefits after 48 hours were stronger for experimental groups under retroactive versus proactive designs when compared to the No Δ condition, which served as a positive control.

Comparisons of the same base pair condition across experiments were performed between the narrower stimulus set, retroactive versus proactive experiments (top left), the wider stimulus set, retroactive versus proactive experiments (top right), the narrower stimulus set, flipped test experiment (in which base pairs were subject to retroactive effects) versus proactive experiment (bottom left), and among the narrower stimulus set retroactive experiment versus flipped experiment (bottom right). In all cases, experimental groups that showed facilitation relative to the Control condition in the overall analyses showed greater retroactive than proactive facilitation.

Wider stimulus set examples and coverage.

(a) Examples of stimuli from the wider stimulus set for all conditions. (b) Variable coverage. Distributions of cue and target GloVe values are shown along the target identity (green) and cue identity (blue) lines, respectively. Distributions of GloVe values are shown as purple points inside the 2-D surface (scatterplot) for the ΔBoth condition.

Supplemental pair memory differed as a function of overall stimulus set relatedness, delay, and word pair condition.

All comparisons were significant except where labeled with gray bars and ‘ns’ (p > 0.1), and data points from individual subjects were jittered slightly for better visualization.

Intrusions did not increase with target relatedness in the ΔTarget condition in any experiment.

We plotted across-subject intrusion rates of reporting supplemental pairs during base pair testing against target relatedness [Top: AS; bottom: GloVe]. There were no positive relationships between intrusions and target relatedness, though there was a significantly negative relationship in the wider stimulus set, 5-min delay experiment.

Bivariate cue and target relatedness had no effect on memorability, though it did significantly predict interdependence in the narrower stimulus set in the 48-hr delay experiment.

These plots are created by putting a white 95th percentile threshold of bootstrapped samples in white grid marks above the purple (true) data, so the presence of purple indicates significance at the p < 0.01 level. (a) Across-subject memorability for each base pair – control was plotted against the bivariate level of cue and target relatedness [top: AS; bottom: GloVe]. (b) Across-subject dependence for each base pair target-supplemental pair target duo was plotted against cue and target relatedness. High levels of cue and target relatedness increased dependence in the narrower stimulus set, 48-hr delay experiment.

Memory and interdependence data from preceding analyses plotted using a common GloVe metric.

(a-b) Across-subject memorability (top) and dependence (bottom) for the ΔTarget (left), ΔCue (middle), and ΔBoth (right) conditions were plotted after combining narrower and wider stimulus sets within a delay. In the 5-min delay experiments (a), we found no significant relationships between relatedness and memorability, though there were relationships between relatedness and interdependence in the ΔTarget and ΔCue conditions. In the 48-hr delay experiments, relatedness predicted memorability in in the ΔTarget and ΔCue conditions, and relatedness predicted interdependence in all conditions. Correlations between the combined correlation are plotted in darker colors, with each individual correlation between stimulus set and relatedness replotted in lighter colors for visual comparison. Text labels reflect correlation values for the combined datasets.

Memory and interdependence data in a paradigm in which we used base pair learning followed by supplemental pair learning and then flipped the order of the tests (supplemental pairs followed by base pairs).

(a) Overall memory performance for base (left) and supplemental pairs (right) by condition. All comparisons were significant in both plots, and data points from individual subjects were jittered slightly for better visualization. (b) Correlations between retroactive memory effects (left) and base-supplemental pair dependence (right) in the ΔTarget (top), and ΔCue (middle) conditions. All correlations between memorability and relatedness were significant, whereas correlations between relatedness and interdependence were significant in the ΔCue and ΔBoth conditions. Pearson correlations are provided within the plots, followed by * when p < 0.05 and ** when p < 0.01.

Direct comparisons of the same base pair condition across 5-minute delay experiments were performed between the narrower stimulus set, retroactive versus proactive experiments (left) and the wider stimulus set, retroactive versus proactive experiments (right).