2.1.2. Task

The task used was designed to interchangeably recruit a rule-based or similarity-based categorization strategy to categorize three categories of football defensive formations. When between-category similarity is low and within-category similarity is high, visual similarity can guide categorization without the need of limited cognitive resources. However, when between-category similarity is relatively high compared to within-category similarity, successful categorization requires the discovery and an application of an explicit categorization rule, which taxes limited resources, such as working memory and attention. Thus, we chose two formation categories that were visually similar to each other and one category that was visually distinct from the other two. For the two similar categories, the subjects needed to discover an explicit counting rule in order to categorize the members of these two groups reliably: One category of formations displayed three people on the line of scrimmage (and four behind them, 3-4 category), while the other had four people on the line of scrimmage (and three behind them, 4-3 category) (Figure 1). Because of the variable starting positions among players, the formations that fell into one category (e.g., 3-4) could look different from each other (within-category variability). In addition, some of the formations in the 3-4 category were visually very similar to some of the formations in the 4-3 category (high between-category similarity). Thus, the participants could not easily rely on visual similarity and instead needed to discover the rule (number of people on the line of scrimmage) that differentiated the categories. This forced subjects to focus their attention to the line of scrimmage while ignoring irrelevant players positioned elsewhere on the field. We expected that, although these were two categories that participants needed to differentiate between, stimuli from both categories would evoke the same cognitive processes (explicit, rule-based categorization). We collapsed over them in neuroimaging analyses when looking for neural processes associated with these rule-based trials, as subsequent analyses confirmed that participants' performance was comparable for the 3-4 and 4-3 categories on all metrics.

**Figure 1.** Standard formations used in the experiment. Left (blue underline): A 6-1 formation, which represents the visually distinct category, had six white players on the line of scrimmage (red line) and one player lined up behind them. Middle and Right (red underline): A 3-4 or 4-3 formation, which represent the visually similar categories, had three players (3-4) or 4 players (4-3) on the line of scrimmage (red line) and either four players (3-4) or 3 players (4-3) lined up behind them. The position of the green players did not vary significantly between formations.

For the visually distinct category, there was also an explicit rule, with six people on the line of scrimmage and one person behind (6-1) (Figure 1). However, this category was visually sufficiently distinct from the 3-4 and 4-3 categories, so subjects could rely on visual similarity alone during categorization, rather than having to discover and invoke an explicit rule.

Every category had three separate formations, each sharing the defining number of players on the line of scrimmage for that category, for a total of nine formations used throughout the experiment. On each training trial, the subjects were shown a random formation for 2.5 s and were tasked with pressing a button on a keypad to place the formation into one of the three categories during the 2 s window (Figure 2). Corrective feedback was given to the subject immediately after making their response and it was on the screen for 1.5 s. The inter-trial-interval was optimized for event-related-design while using Optseq2 software and varied between 2 and 8 s [24]. Each formation was shown six times during each training block and there were six total training blocks.

A generalization block was implemented at the end of the experiment in order to test the subject's ability to generalize the strategies that they acquired during training. During this block, the nine old formations were intermixed with nine new formations that belonged to the learned category structures. Each stimulus was randomly shown one at a time and was on the screen for 2 s while the subject used a response pad to categorize the stimulus. No feedback was given during this block, and instead a black fixation screen was shown for 10 s before the next stimulus was presented—resulting in a total fixed trial length of 12 s (Figure 3). Each new and old stimulus was shown only once during the generalization block. The inclusion of a generalization block allowed for us to test whether the participants indeed discovered the category structure rather than memorized labels for individual examples, as rote memorization during training would hinder their categorization performance on the new formations.

**Figure 2.** Formations were shown for 2.5 s. Immediately following a response, contingent feedback was shown for 1.5 s. Upon feedback termination, a fixation mark was shown for the duration of the inter-trial interval of 2–8 s before the next formation was presented.
