Fig 1.
Task design and behavioral results.
(A) Two arrays were presented sequentially in each trial. In half the trials, two gratings were presented serially (sample 1 and sample 2). In the other half of trials, a single grating was presented either at the first (25% of trials) or second (25% of trials) time point. A color cue presented at the same time as the second sample indicated which grating the participant should recall. When the probe appeared, participants rotated it to match the orientation of the target item using button presses. The report was registered by pressing a third button. A grating with the correct orientation was presented as feedback. (B) Behavioral serial dependence induced by sample 1 or previous target. The mean signed error of recall errors as a function of the relative orientation of the inducer to the current target. Data was smoothed for visualization purpose only. The smoothing was done by binning the relative orientation of the inducer into 64 evenly spaced, overlapping bins, with each bin containing the 25% of trials closest to its bin center. Shading indicates SEM.
Fig 2.
Decoding of the current target and the neural bias on the target induced by sample 1 or the previous target.
(A) Orientation decoding of the current target. Decoding evidence of zero would suggest no successful decoding. Red horizontal lines indicate timepoints with significant decoding assessed with cluster-based permutation tests. The strong decoding of target after the response was likely driven by the feedback stimulus showing the correct orientation of the target, which was presented 200 ms after the response. Gray shading indicates SEM. The plot below shows the distribution of response times (estimated with the time when participants submitted their reports) across participants and trials. Ninety-nine percent of the reports were made 700–6,200 ms after probe onset. Note that in all probe-locked analyses, the later part of the epoch (while and after most responses were given, starting at approximately 2,000 ms) contained increasing amounts of noise from irrelevant signals after the current recall period. (B) Bias of the neural representation of the current target caused by the previous target. Trials were sorted into two groups according to whether its previous target is clockwise (red line) or counterclockwise (blue line) to the current target. The y-axis shows the mean asymmetry index, with a positive number indicating a clockwise-biased neural representation. A smoothed time course is also shown on top (bold lines) for visualization purpose only. The smoothing was done by convolving a Gaussian filter (s.d. = 80 ms) with the raw asymmetry index time course (thin lines). Shadings around the thin lines indicate SEM. The gray-shaded regions indicate time windows with which paired t-tests were conducted. (C) Bias of the neural representation of the current target caused by sample 1, with the same plotting conventions as in B. (D) Tuning curves (mean-centered within each participant) showing the multivariate reconstruction of the current target, in clockwise-previous-target trials (red) and counterclockwise-previous-target trials (blue) separately, averaged in the last 2 s before response. Shadings indicate SEM. CCW, counterclockwise; CW, clockwise. (E) Tuning curves showing the reconstruction of the current target, in clockwise-sample-1 trials (red) and counterclockwise-sample-1 trials (blue) separately, averaged in the time window from 254 to 834 ms after probe onset.
Fig 3.
The MEG topographies showing which sensors contributed more to the effects.
(A) The topography of the sample-1-induced repulsive bias, averaged in the time window from 254 to 834 ms after probe onset. The bias index was calculated by subtracting the asymmetry index of counterclockwise inducer trials from clockwise inducer trials. A positive bias index represents an attractive neural bias. (B) The topography showing which sensors contributed the most to the decoding of target in the same time window. (C) The topography of the previous-target-induced attractive bias in the last 2 s before response. (D) The topography of target decoding in the same time window.
Fig 4.
Asymmetry index of the representation of the current target.
(A) For reference, the neural bias of the target induced by the previous target (identical to Fig 2B). (B) When trials are sorted according to the relative orientation of the participant’s report in the current trial, there is no bias. Plotting conventions are the same as in Fig 2.
Fig 5.
Decoding of the target from the previous trial, with the same plotting conventions as in Fig 2A.
Since no significant period of decoding was found with cluster-based permutation test, we also applied one-sample t-tests in the time window where a previous-target-induced attractive bias was found (gray-shaded region). No successful decoding was found in this time window either.
Fig 6.
Task design and behavioral results of the EEG dataset.
(A) Two samples were presented sequentially in each trial. An auditory retrocue indicated which grating the participant should recall. Then, the probe appeared with a random starting orientation, and participants rotated it to match the orientation of the target item using button presses. The report was registered by pressing “Enter”. Feedback was provided with two small light gray discs presented at the edge of the grating to indicate an imaginary line corresponding to the correct orientation of the cued item. (B) Behavioral serial dependence induced by sample 1 or previous target. The mean signed error of recall errors as a function of the relative orientation of the inducer to the current target. Data was smoothed for visualization purpose only. The smoothing was done by binning the relative orientation of the inducer into 64 evenly spaced, overlapping bins, with each bin containing the 25% of trials closest to its bin center. Shading indicates SEM.
Fig 7.
Decoding of sample 1, sample 2, and target of the current trial.
(A) Orientation decoding of sample 1 (purple curve) and sample 2 (orange curve). Decoding evidence of zero would suggest no successful decoding. Horizontal lines indicate timepoints with significant decoding assessed with cluster-based permutation tests. Shading indicates SEM. The inset plot shows the distribution of the time when participants started rotating the probe (see Methods). Across participants and trials, 99% of rotation-start times fell into the range of 635 to 5,367 ms. 91.70% of rotation-start times were shorter than 3 s. The dotted vertical lines indicate the time of sample 1 onset (0 ms), sample 2 onset (750 ms), retrocue onset (1,500 ms), and probe onset (3,000 ms), respectively. (B) Orientation decoding of the target.
Fig 8.
The neural bias induced by the previous target (first row) and sample 1 (second row) during recall of the current target.
(A) Bias of the neural representation of the current target caused by the previous target. Trials were sorted according to whether the previous target was clockwise (red line) or counterclockwise (blue line) to the current target. The y-axis shows the mean asymmetry index, with a positive number indicating a clockwise-biased neural representation. Shadings around the thin lines indicate SEM. The gray-shaded regions indicate time windows when sample decoding was significant and over which data were averaged for statistical inference. (B) Bias of the neural representation of the current target (i.e., sample 2) caused by sample 1 in sample-2-cued trials. Trials were sorted according to the orientation of sample 1 relative to the current target.
Fig 9.
Asymmetry index of the representation of the current target when trials were sorted according to the relative orientation of the participant’s report in the current trial.
Plotting conventions are the same as in Fig 8.
Fig 10.
Decoding of the target from the previous trial, with the same plotting conventions as in Fig 7.
The gray-shaded regions indicate time windows in which one-sample t-tests were conducted.