Auditory-motor adaptation and de-adaptation for speech depend more on time in the new environment than on the amount of practice

Abstract

Sensorimotor adaptation is critical for learning and refining voluntary movements. One common assumption is that the number of practice trials fully determines the amount of adaptation. It is possible, however, that for some tasks the sensorimotor system continues to learn during the time in-between executed movements as long as there is no evidence that the environment has changed. The amount of time spent in the altered environment (total exposure time) then would be more important than the number of practice movements performed during that time. In the current study, we investigated adaptation and de-adaptation as a function of practice trials versus exposure time using speech articulation as the model system. Four separate groups of 14 participants read out loud monosyllabic words at a rate of either 18 words per minute or only 6 words per minute during the adaptation and de-adaptation phases of a speaking task with formant-shifted auditory feedback. The data demonstrate that both auditory-motor adaptation and de-adaptation depend more on exposure time than amount of practice. COIN model simulations suggest that this common effect is consistent with de-adaptation constituting active re-learning of the unaltered environment rather than forgetting of the learned behavior.

Fig. 1. Methodological information.

A Schematic representation of the experimental setup. B Example spectrogram of a single trial for the target word “Tuck” with first (F1) and second (F2) formant frequency tracks for the vowel selection superimposed (plum = participant acoustic output, blue = effects of +250 cents shift perturbation heard in the auditory feedback). C Practice schedules for four experimental groups who experienced different combinations of intensive and non-intensive trial presentation in the adaptation and de-adaptation phases. All groups completed the same number of trials in each phase: 54 trials (18 of each word) baseline, 108 trials (36 per word) adaptation, and 108 trials (36 per word) de-adaptation. ITI inter-trial interval.

Fig. 2. A non-intensive practice schedule increased the amount of both adaptation and de-adaptation.

Shaded areas indicate 1 standard error for each group (n = 28 participants per group). A Group average formant frequencies, after individual normalization relative to baseline, for Intensive Adaptation (green) and Non-Intensive Adaptation (orange), expressed by time spent in the perturbation phase. The Intensive Adaptation group completed all 36 perturbed epochs (3 trials per epoch) in 6 min, whereas the Non-Intensive Adaptation group completed these same epochs in 18 min. Both groups achieved a similar extent of learning after 6 min of perturbation (despite the different number of practice trials), but the Non-Intensive Adaptation group continued to adapt more during the remaining time. B When comparing the same data by number of epochs, the Non-Intensive Adaptation group showed a larger final extent of learning (last three epochs). C Group average formant frequencies, after individual normalization to the end of the perturbation phase, for Intensive De-Adaptation (light blue) and Non-Intensive De-Adaptation (plum) across the after-effects phase. After 6 min, when the Intensive De-Adaptation group had completed all 36 epochs (3 trials per epoch), both groups achieved a similar extent of de-adaptation, but the Non-Intensive De-Adaptation group continued to de-adapt during the remaining time. D When comparing the same data by number of epochs, the Non-Intensive De-Adaptation group showed a larger final extent of de-adaptation (last three epochs).

Fig. 3. Two-rate state-space model fitting for speech auditory-motor adaptation data.

Left: The state-space model simulations (lines) generated by the average parameter values of 5000 bootstrapped model parameters show good fits to the group average empirical data (dots). Right: The Non-Intensive Adaptation group showed larger slow retention factors than the Intensive Adaptation group as determined by 95% percentile confidence intervals.

Fig. 4. Two-rate state-space model fitting for groups with Intensive vs. Non-Intensive practice schedules when completing speech auditory-motor de-adaptation.

Left: The average state-space model parameters used for the adaptation data—with a larger slow retention factor ($A_s$) for Non-Intensive practice than for Intensive practice—generated simulations (lines) that did not fit the empirical data for de-adaptation (dots). Middle: Swapping the slow retention factors between the de-adaptation two groups’ models results in simulations that do fit the empirical data. Right: The COIN model simulation with a large prior mean of retention (${\mu }_a$) was similar to the Non-Intensive de-adaptation data, whereas the simulation with a smaller prior mean of retention was similar to the Intensive de-adaptation data.

Fig. 5. Adaptation and (inverted) de-adaptation profiles were similar (n = 28 participants per group).

After the de-adaptation data were normalized to a new baseline (i.e., starting near 0 cents) and vertically inverted (i.e., changing in the same direction as adaptation), the resulting de-adaptation profile was similar to Intensive De-Adaptation. Shaded areas indicate 1 standard error for each group.