Perceptual learning evidence for an interval- and modality-invariant representation of subsecond time

A central theme in time perception research is whether subsecond timing relies on a dedicated centralized clock, or on distributed neural temporal dynamics. A fundamental constraint is the interval- and modality-specificity in perceptual learning of temporal interval discrimination (TID), which argues against a dedicated centralized clock, but is more consistent with multiple distributed mechanisms. Here we demonstrated an abstract, interval- and modality-invariant, representation of subsecond time in the brain. Participants practiced TID at a specific interval (100 ms), and received exposure to a transfer interval (200 ms), or to a different auditory/visual modality, through training of an orthogonal task. This double training enabled complete transfer of TID learning to the untrained interval, and mutual complete transfer between visual and auditory modalities. These results demonstrate an interval- and modality-invariant representation of subsecond time, which resembles a centralized clock, on top of the known distributed timing mechanisms and their readout and integration.

The auditory TID training also improved visual TID at the same 100-ms interval, 3 reducing visual TID thresholds by 0.14 ± 0.04 log units (t = 2.56, p = 0.012, Cohen's d = 0.90, 4 95% CI [0.03, 0.24]) (Fig. 1a, b). However, for the visual TID training baseline group,  (Fig. 1c, d). These data thus replicated the asymmetric transfer of TID 9 learning from auditory to visual, but not the other way around. Here the visual TID 10 improvement through auditory TID training (V100 in Fig. 1b) was about 70% of that through 11 direct visual TID training (V100 in Fig. 1d), suggesting that auditory TID training might have 12 not maximized the visual TID performance in these observers. interval discrimination (TID) learning. a. The mean auditory TID learning curve at a 100-ms 3 interval (A100), and pre-and post-training auditory TID thresholds at a 200-ms interval (A200) 4 and visual TID thresholds at a 100-ms interval (V100). b. The mean improvements of TID with 5 trained A100, untrained A200, and untrained V100 conditions. c. The mean visual TID learning 6 curve at a 100-ms interval (V100), and the pre-and post-training auditory TID thresholds the 7 same interval (A100). d. The mean improvements of TID with trained V100 and untrained A100. Here and in later figures, error bars indicate ±1 standard error of the mean, and solid and empty 1 histograms indicate threshold improvements from training and transfer, respectively. 2 3 Complete transfer of TID learning across intervals with double training 4 We then investigated whether double training could abolish interval specificity, a major 5 constraint on modeling the mechanisms of subsecond time perception. Eight participants 6 practiced auditory TID at a 100-ms interval, as well as an orthogonal tone frequency 7 discrimination task at a 200-ms interval, in alternating blocks of trials for five sessions. Our 8 hypothesis was that this secondary frequency discrimination training would allow the 9 participants to receive exposure to the 200-ms interval, so as to promote TID learning transfer 10 from a 100-ms interval to a 200-ms interval.

11
Double training improved auditory TID at a 100-ms interval by 0.27 ± 0.07 log units (t = 12 5.15, p < 0.001, Cohen's d = 1.82, 95% CI [0.17, 0.38]), which was comparable to the 0.29 log 13 unit improvement in baseline training (A100 in Fig. 1b Fig. 2c), which was significantly higher than the 2 corresponding 0.08 log-unit improvement in the baseline condition (A200 in Fig. 1b)  learning transfer was complete, all participants continued to practice the auditory TID task at a  To ensure that the learning transfer effect was not due to the tone frequency discrimination 10 training per se, a control group only practiced tone frequency discrimination at a 200-ms 11 interval. The training improved frequency discrimination by 0.14 ± 0.04 log units (t = 2.59, p =  We first tested whether double training could enable transfer of TID learning from 10 visual to auditory. Nine participants practiced visual TID at a 100-ms interval. They also 11 received exposure to the auditory 100-ms interval by practicing an orthogonal tone frequency  interval. c. The mean improvements of visual and auditory TID thresholds (V100 & A100) and 1 tone frequency discrimination thresholds (FD100) in double training and control conditions, as 2 well as in the earlier baseline condition (replotted from Fig. 1d). 3 We were also curious whether double training could lead to better auditory-to-visual 4 TID learning transfer than the baseline auditory TID training only condition, in which the 5 visual TID improvement from auditory TID training was about 70% of that from direct visual 6 TID training (Fig. 1b, d). Eight new participants practiced auditory TID and visual contrast 7 discrimination, both at a 100-ms interval, in alternating blocks of trials in the same training  (Fig. 4a, c). Planned comparisons showed that the visual TID 13 improvement was significantly higher than the 0.14 log-unit improvement in the auditory TID  assumptions for some centralized timing mechanisms that can process multiple 7 durations/intervals, without caring too much about these early models' specifics. These 8 centralized mechanisms or the clock may represent different durations/intervals in an abstract 9 manner (e.g., through standardization of incoming neural inputs representing different 10 durations/intervals). Moreover, perceptual learning may improve the precision of this abstract 11 time representation, so that TID learning can transfer from a trained interval to a new interval. 12 The asymmetric transfer of TID learning from audition to vision has been interpreted as 13 auditory timing mechanisms, which may have less internal noise and thus be more precise, for an abstract representation of subsecond time, which is independent of not only specific 7 intervals, but also sensory modalities and timing precisions.  The stimuli for tone frequency discrimination were the same as those for auditory temporal 5 interval discrimination, excepted that the frequencies of two pairs of pips were changed while 6 the temporal intervals were fixed. The two pairs of pips, one pair at a standard frequency of 1 7 kHz and the other at a higher comparison frequency (1 kHz + Δf), were presented subsequently 8 in a random order in each trial. The participants pressed the left or right arrows to indicate 9 whether the first or second pair of tone pips had a higher frequency. Happy or sad cartoon face 10 was provided as feedback.

11
The frequency discrimination thresholds were measured with a 2IFC staircase procedure. 12 The starting frequency difference (Δf) between the standard and comparison stimuli was 50%, 13 which decreased by a factor of 2 after every correct response until the first incorrect response. 14 rate. Each staircase ended after 60 trials. The threshold was calculated as the mean of the last 1 40 trials.

2
The stimuli used for visual contrast discrimination was the same as those for visual 3 temporal interval discrimination, except that the Gabor contrast was varied while the interval 4 was fixed (100 ms). Only one pair of Gabor was presented in each trial. In 80% of the trials, the 5 two Gabors had identical contrast, which randomized from 0.15 to 1. In the remaining 20% 6 trials, the contrasts of the two Gabors differed by 50%. The participants judged whether the two 7 Gabors had identical contrast. Happy or sad cartoon face was provided as feedback. The d' 8 values were calculated to measure the contrast discrimination performance.

9
Each experiment consisted of a pre-training session, five training sessions, and a post-10 training session on separate days. The experiment was completed within 7-13 days, with inter-11 session gaps of no more than 2 days. Each single-training session consisted of 16 blocks of 12 trials and lasted for approximately 1.5 hours. Each double-training session consisted of 10 13 blocks of trials for the primary task and 10 blocks of trials for the secondary task in an 14 alternating order, and lasted for approximately 2 hours.
The sample size was decided on the basis of a previous temporal interval 1 discrimination learning study that used similar stimuli (Fig. 4, 100   was performed by Linear Mixed Effects (LME) modeling to examine the training and transfer 10 effects of the TID task, using the "lmer" function from the "lme4" package (Pinheiro & Bates, 11 2000). All groups' data were included in a single LME model to reduce Type-I error. The TID 12 thresholds were first log-transformed to achieve normal distributions (Shapiro-Wilk test 13 before log-transformation: p < 0.001 for auditory TID thresholds at 100 ms and 200 ms 14 intervals, and visual TID thresholds at 100 ms intervals; Shapiro-Wilk test after log-  The LME analysis revealed significant main effect of Test (F(1, 55) = 68.71, p < 9 0.001) and Condition (F(2,28) = 86.72, p < 0.001), but not Group (F(7,44) = 0.18, p = 0.99).