This proposal intends to tackle a fundamental issue of problem solving in education, by investigating the extent to which learners’ ability to consider a given problem from different perspectives depends on the cognitive load of the situation. Recategorizing a problem by constructing a new representation of the situation is known to be a difficult endeavor relying on cognitive flexibility (Clément, 2022; Gvozdic et Sander, 2020). One of the reasons why this process is so challenging is that going beyond the initial semantic encoding of a word problem requires to detach from its most superficial features and to construct a new encoding, closer to the problem’s deep structure (Gros, Thibaut, & Sander, 2020). This process, termed semantic recoding, has been described as cognitively demanding, being associated with lower performance rates, longer response times, and greater pupil dilation (Gros, Sander, & Thibaut, 2019; Gros, Sander, & Thibaut, 2020). According to Cognitive Load Theory (Sweller, 2015, 2016), helping learners direct an optimal amount of cognitive resources toward activities that are directly relevant to learning should facilitate performance and transfer (Chandler & Sweller, 2009). Conversely, increasing the cognitive load attached to a task may significantly hinder participants’ performance on said task (Sweller, Van Merrienboer & Paas, 2019; Puma & Tricot, 2019). Since semantic recoding has a non-negligible cognitive cost, problems requiring that learners engage in such a change of perspective should especially benefit from the alleviation of the cognitive load attached to the situation. On the other hand, performance on problems that do not require a semantic recoding should be less impacted by an increase in cognitive load.To investigate this hypothesis, we designed 2 experiments using 12 isomorphic word problems, half of which required a semantic recoding to be solved in 1-step. We aimed at evaluating the extent to which participants’ recoding of the situation can be fostered or hindered by variations in the cognitive load attached to the problems, be it their intrinsic or extraneous load.In the first experiment (N = 145), we used a dual task paradigm and asked participants to solve the problems in the least number of steps. Each problem was coupled with a dot memory task of varying difficulty. We expected participants to experience more difficulty in recoding the problems when performing a high-load dot memory task, as compared to problems coupled with a low-load dot memory task. Preliminary results suggest that, contrary to our hypothesis, participants’ performance was mostly guided by the problems’ semantic content, which did not interact with the extraneous cognitive load attached to the problems. In a second experiment (N = 40), we varied the intrinsic difficulty of the problems instead, by using either low-range or high-range numbers to alter the difficulty of the calculations to solve the problems. Preliminary results suggest that, similarly to Experiment 1, the increase in difficulty affected all problems in a similar fashion, regardless of whether they required a semantic recoding. We discuss the potential limitations of these preliminary studies and propose next steps to further investigate the interactions between cognitive flexibility, cognitive load, and content effects in problem solving.