Auditory category learning is robust across training regimes (Obasih et al., 2023)

Open Materials Overview:

This Open Materials page gathers all of the materials used to the run the online experiments described in the journal article "Auditory category learning is robust across training regimes" (Obasih et al., 2023). The study comprised of five experimental conditions that manipulated the category learning training regime to assess differential effects on learning and generalization of complex, auditory non-speech categories.

For each experimental condition, we have included here the full Gorilla Experiment (GExperiment) for each of the five experimental conditions. For each full GExperiment, participants signed a consent form, filled out a demographics form, performed a volume and headphone check (Milne et al., 2021 - linked in the materials below), and then after the main experiment, filled out an exit questionnaire that also contained the experiment debriefing. The Gorilla Questionnaires (GQuestionnaire) for the demographics form and exit questionnaires are included below. These were identical across experimental conditions.

For each experimental condition, we have also included the Instructions, which were GTasks separate from the main experiment, as well as the GTask for the main experiment. For every experiment, the GTask comprised one block, which was repeated four times in the GExperiment tree for a total of four blocks.

The research data set and analysis files can be found here: https://osf.io/bnk9e/

Task Stimuli Overview:

For each experiment, participants were tasked with learning four auditory categories. The four categories were created from non-speech analogous of speech recordings of word-frames spoken with each of the four lexical tones in Mandarin (thus each category was an analog of one of the four tones). However, to increase the complexity of the stimulus space, each exemplar contained two simultaneous frequency bands (high or low), each with three non-speech hums. One of frequency bands contained a regular pattern of three hums of the same lexical tone contour (thus, the informative band), while the other band contained an irregular pattern of three hums, one each from any of the four lexical tone contours (thus, the distractor band).

In one stimulus condition, the informative band contained a regular pattern of hums from Tones 1 and 2 in the high frequency band (Categories A and B, respectively), while the informative band contained a regular pattern of hums from Tones 3 and 4 in the low frequency band (Categories C and D, respectively). In the other stimulus condition Categories A and B were respectively defined by an informative band containing a regular pattern of hums from Tones 1 and 2 in the low frequency band, while Categories C and D were respectively defined by an informative band containing a regular pattern of hums from Tones 3 and 4 in the high frequency band. The two stimulus conditions were for counterbalancing purposes.

There were 120 training trials 20 generalization trials per block, and thus 480 training trials and 80 generalization trials per experiment. The training trials were randomly selected from a pool of 512 training stimuli while the generalization trials were pseudo-randomly selected pre-experimentation from a separate pool of 512 novel generalization stimuli. The pool of 512 training stimuli and 80 novel generalization stimuli were consistent across experimental conditions.

Task Procedure Overview:

For each experiment, the general task was the same. One block comprised of a training block and a testing block. The participant heard a sound while looking at four alien creatures on the screen and were instructed to learn which sound belonged to which alien (categorization task). During each training trial, the participant was presented with the auditory stimulus exemplar, asked their answer with a keyboard button press, and given feedback on the correctness of their answer. After the training block, during each testing trial, the participant was presented with a novel exemplar not heard during training, asked to record their answer with a keyboard button press, and NOT given feedback (generalization task). After a short break, the task commenced to the next block, until four blocks were completed.

Each block (training block + generalization block) takes about 10 minutes to complete. Four blocks of the task takes about 45 minutes to complete. The full experiment, from consent to debriefing, takes about 60 minutes to complete.

Experimental Conditions Overview:

The conditions differed by training regime as follows:

Experiment 1 (4AFC Full Variability): Training trials were 4-alternative forced choice (4AFC). Each block was broken up into three mini-blocks of 40 trials each (120 trials/block). The training exemplars were pulled from the full pool of 512 training stimuli.

Experiment 2 (4AFC Low Variability): Training trials were also 4AFC and training blocks were broken up into three mini-blocks of 40 trials each (120 trials/block). However, the training exemplar were reduced to a pool of 40 exemplar pseudo-randomly chosen pre-experimentation from the original full pool of 512 training stimuli. Thus, for this experiment, the 40 training exemplars were repeated each training mini-block for training block, in randomized order.

Experiment 3 (2AFC Grouped Pairs by Informative Band): Training trials were instead 2-alternative forced choice (2AFC), where the participant was instructed to pick between two aliens on screen while the other two were greyed out. For this experiment, the two categories to choose between were deliberately paired based on sharing the same informative frequency band (high or low). Here, each block was broken up into six mini-blocks of 20 trials each (120 trials/block), with mini-blocks alternating presentation of Categories AvB and Categories CvsD three times (i.e., the mini-blocks were AB, CD, AB, CD, AB, CD or CD, AB, CD, AB, CD, AB; these two ordering conditions were for counterbalancing purposes). Like Experiment 1, the training exemplars were pulled from the full pool of 512 training stimuli.

Experiment 4 (2AFC Random Pairs): Training trials were also 2AFC and training blocks were broken up into six mini-blocks of 20 trials each (120 trials/block). However, this time the two categories to choose between were randomly paired regardless of sharing informative frequency band or not. All possible pairs were presented during the six mini-blocks per training block (AB, AC, AD, BC, BD, CD), however, the order of the mini-blocks was random from block to block and participant to participant. Like Experiments 1 and 3, the training exemplars were pulled from the full pool of 512 training stimuli.

Experiment 5 (2AFC Grouped Pairs by Informative Band with Explicit Instructions): This experiment was identical to Experiment 3 except that in the instructions, participants were informed that "previous participants of the experiment have found it helpful to listen to the higher or lower pitched sounds" and were presented with a screen instructing them to "Listen high!" or "Listen low!" in accordance with the informative frequency band of the categories in the upcoming mini-block. As with Experiment 3, each block was broken up into six mini-blocks of 20 trials each (120 trials/block), with mini-blocks alternating between AvsB and CvsD (two counterbalanced ordering conditions: AB, CD, AB, CD, AB, CD or CD, AB, CD, AB, CD, AB). Like Experiments 1, 3, and 4, the training exemplars were pulled from the full pool of 512 training stimuli.

Generalization Trials:

In every experimental condition, regardless of if the training regime included 4AFC or 2AFC trials, the novel generalization blocks were 4AFC trials. Each generalization block was a single block of 20 trials, thus 80 trials across the whole experiment. The 80 novel generalization exemplars were identical across experiments, randomized in order between participants. This allowed us to directly compare the effects of training regime on generalization ability.

Using These Materials

If you would like to use these materials and find them useful in your work, please cite the article:

Obasih, C. O., Luthra, S., Dick, F., & Holt, L. L. (2023). Auditory category learning is robust across training regimes. Cognition, 237, 105467. https://doi.org/10.1016/j.cognition.2023.105467

And please cite or link to the current open materials page: https://app.gorilla.sc/openmaterials/462043

Back to Open Materials