Milgram and Kishino (1994) defined AR as a system that makes it possible to superimpose a virtual object on a real image. AR takes place in a Reality–Virtuality continuum (Figure 1), which is very close to the real environment side and can be considered as a composite view. On the opposite continuum side, the virtual environment is a computer-generated environment (which is also called virtual reality).

One of the main objectives of AR is to allow clear connections to be made between the real-world and abstract scientific concepts (Laine et al., 2016). AR allows natural interactions between students and learning objects, encouraging the creation of embodied representations of learning concepts. The cognitive dimension helps students to better symbolically understand the scaffolding of learning progression. This dimension is made possible thanks to the spatio-temporal alignment of information. Thus, the cognitive dimension improves the understanding of abstract concepts (Bujak et al., 2013). Moreover, AR creates opportunities for collaborative learning around virtual content and in a non-traditional environment (Bujak et al., 2013; Martín-Gutiérrez et al., 2015; Peramunugamage et al., 2023). More recently, Chang et al. (2018) argued that AR technology can be used to teach scientific reasoning and promotes students’ positive attitudes toward scientific issues. AR improves students’ reflective skills in science lessons (Arici and Yilmaz, 2023). AR can also promote student autonomy by allowing an adaptation of the content for each student. Hence, teachers can adapt the support according to the learners’ demands and constraints, and their different learning processes (Amadieu and Tricot, 2014).

Several studies have highlighted the strong limitations of AR in teaching. For instance, Huang et al. (2016) and Wang (2017) showed that the students’ learning performances are better when AR is coupled to more traditional learning support. Furthermore, a clear negative impact of AR has been also highlighted. Some studies that have used a new learning environment have found that the AR-based learning approach fostered the cognitive overload of students (Van Bruggen et al., 2002; Hron and Friedrich, 2003; Ali et al., 2022). Indeed, cognitive overload corresponds to a mental state in which a student is engaged in a task that is very demanding. In this case, the student does not have sufficient cognitive resources to perform the task easily. This demanding cognitive task requires the student to devote a great deal of attention, to take longer to complete a task with risk of making a lot of mistakes (Tricot, 2023). AR could be considered as a deleterious tool because of the cognitive overload generated using this new technology related to the complex tasks they perform (Wu et al., 2013). According to Dugas (2018), some elements still need to be improved before the generalization of this type of technology. The hardware may have some technical limitations, or the interface may lack ergonomics or adaptation that can sometimes lead to visual and cognitive overload. The way the pictures are displayed (arrangement in the operator’s visual field, ergonomics of the symbols, etc.) is crucial to achieve efficient performance for learners. If the technology mediates too much or too less information, the performance of the participants could be impacted (Alexander et al., 2005). Students can be cognitively overload by the related learning-environment information and by the complexity of the tasks that they have to complete (Wu et al., 2013; Ali et al., 2022; Buchner et al., 2022). Nevertheless; the use of guidance in AR animations is important for increasing students learning and reducing their cognitive load (Yilmaz, 2023). Other findings have outlined that AR learning can generate attention issues, showing that the learners’ attention is more focused on the AR system than on the content knowledge (Tang et al., 2003; Morrison et al., 2009; Radu, 2014).

Many researchers consider that mental effort is a good indicator for tracking cognitive load (Paas and Van Merriënboer, 1994; Antonenko and Niederhauser, 2010; Haji et al., 2015; Yilmaz, 2023). Mental effort and cognitive load are two close concepts that are rooted in Sweller’s cognitive load theory Sweller (1988). The objective of this theory is to understand the mechanisms that are involved in school performance differences in learning tasks. According to the classical architectures of the cognitive system, cognitive load theory describes how the limited resources of working memory are used when the cognitive system is in a learning situation (Sweller, 1988; Kirschner, 2002; Chen et al., 2018). Sweller (1988) distinguishes three cognitive load plans: first, the unnecessary extrinsic cognitive load corresponds to the organization of learning materials; second, the intrinsic cognitive load corresponds to the interaction between the cognitive task to be performed and the hardware to be used; and finally, the relevant cognitive load corresponds to the efforts invested in the construction of cognitive schemas. Although the assessment of mental effort fails to distinguish Sweller’s three dimensions, this kind of subjective rating scale is efficient for reflecting general cognitive processes (Ouwehand et al., 2021).

The inquiry-based learning is defined as a process who highlights new causals relations that can be likened to problem solving. The students formulating hypothesis and testing them before concluding (Pedaste et al., 2015). It involves students in a usually complex process of scientific discovery, in a process divided into logically connected parts, allowing the student to be guided through the main features of the method. These parts are called inquiry phases and together they form an inquiry cycle (Pedaste et al., 2015).

Based on a metanalysis of 32 articles, Pedaste et al. (2015) analyzed the inquiry processes addressed in the scientific literature. Despite the variety of the inquiry-based learning approach, they built an inquiry-based learning framework including five general inquiry phases common to the majority of the analyzed articles. This framework is seen as a circle including many relationships and feedback loops between the different phases. However, it can be summarized as follow:

One of the reasons why inquiry-based learning is increasingly present in education is that its success can be greatly enhanced by recent technical developments that allow the enquiry process to be supported by electronic learning environments (Pedaste et al., 2015). In our context, the French school curriculum promotes the inquiry approach. The teaching scenario proposed in this study is based on the inquiry model drawn up by Pedaste et al. (2015) and fits with the recommendation of the French school institution.

Despite the large number of studies on the effect of AR on mental effort and learner performance, to our knowledge very few of the study has investigated the impact of AR modeling activities in inquiry-based approach. Hence, this study aims to contribute to the exploration of the AR technology’s effectiveness on mental effort and learning outcomes in this specific educational approach. In particular, this study attempts to answer the following questions:

1.To what extent is mental effort impacted by AR during inquiry-based science training (about seasonal phenomena)?

2.How does AR impact the students short-term and long-term learning outcomes?

3.How does mental effort during learning tasks influence scientific learning outcomes?

According to the literature review, AR technology was mainly reported as beneficial for the learner’s mental effort and learning outcomes, even if there is a lack of knowledge about the specific case of AR modeling activities during inquiry-based training. Moreover, an alternative hypothesis emerged from the previous studies, as follows: when technology is used by the learners for the first time, then the learning effect could be limited or may even be negatively impacted.

An experimental protocol has been set up to address the three research questions. The sample consisted of 33 first year Master’s degree students (27 years old, ± 7.5 on average) from the Aix-Marseille University. The students were all engaged in this Master’s degree to be primary school teachers. They are all involved in a training program devoted to the development of their teaching skills and scientific knowledge through inquiry approaches. And all the students agreed to take part in the research process. Since the research objectives fit with training program objectives, the experiment took place during a usual face-to-face science lesson for 90 min and the training was animated by the usual science trainer.

The students were randomly assigned to one of two groups. The control group (n = 17) learned with different physical modeling technologies (e.g., terrestrial globes, torch light, diagrams, pictures), while the experimental group (n = 16) learned with the AR-learning system (i.e., an interactive 3D representation of the Earth-Sun system). The experimental group used only the mobile application ARTEfac (AR for Teachers Education Faculties) to learn the scientific concepts. ARTEfac (Aix-Marseille-Université, 2020) is a free mobile learning app developed by Aix-Marseille University where it is possible for trainers to create simple multimedia mobile learning scenarios or to customize already built-in 3D interactive models developed by researchers, university science trainers and programmers (for more information see: Mascret et al., 2023). The Earth-Sun AR model is one of the already built-in models that we have designed and integrated to the experimental learning scenario. Following the suggested affordances by the literature (Wu et al., 2013; Pedaste et al., 2020), the Earth-Sun AR model was used at the investigation phase and was thought for exploring the model with the possibility to change parameters and observe the consequences on the Earth-Sun system (for instance change the Earth’s tilt).

The researchers and trainer paid particular attention to provide at maximum the same information to the experimental and control groups during the training session. In this way, the trainer (who was the same in both groups, both experimental and control) began by orally announcing the instructions and objectives of the training session before starting. The students were led to make the link between phenomena observed on the Earth and the different parameters, such as the Earth’s rotation and revolution through an inquiry scenario.

More precisely, the training session has been designed on the inquiry leaning principles, respecting the five inquiry phases (Pedaste et al., 2015). During the Orientation phase, the trainer introduced the topic by using images and graphs about the variation of temperature in Europe. The conceptualization was led by the trainer fostering the setup of learners’ inquiry questions and hypothesis. It was the time of the emergence of very common misconceptions such as: the seasons are controlled by the distance between the earth and the sun with the earth farther from the sun in winter than in summer. The Investigation phase was mainly based on the physical or AR modeling in order to explore, observe and interpret the evidences and scaffolded by the worksheet. The only difference between the control and experimental group approaches is rooted in the investigation phase using either the physical modeling or the AR modeling activities. The Conclusion phase was focused on the learners’ main knowledge outcomes about the seasonal phenomenon that they have to retain. The discussions within and between the groups of students were held all along the previous phases. Students communicated and justified their ideas/findings/decisions with the help of the trainer all along the phases of the training session.

The training session lasted 90 and 10 min were necessary to fill-in the retention test 30 days after the training session. At the very beginning of the training session, the students started by filling-in a pre-test to assess their knowledge of the subject. They then participated in the activities about the Earth-Sun system. Finally, they had to complete a short-term test just after the training session and a long-term test 30 days after the session (all were similar to the pre-test). The experimental protocol is illustrated in Figure 5.

Each task, guided by the worksheet, was identical for the control and the experimental group and was evaluated through the mental effort test that was filled-in individually. During the training session, the students were grouped by four or five individuals using one or two tablets in each group. The trainer moved between the groups to ensure a good understanding of the given tasks. The control group used hands-on materials, whereas the experimental group, were introduced to manipulate the AR model. To guide students across the understanding of the Earth-Sun system, they had to complete 10 different tasks (identical for control and experimental groups) that were described in the worksheet. This study was approved by the ethics committee of Aix-Marseille Université (no: 2022-03-10-003).

The students used the proposed AR model for the very first time during this study.

Quantitative dependent variables such as the mental effort and learning outcomes that were obtained in the pre-test, short-term test and long-term tests do not meet the assumption of normality (Shapiro–Wilk test; p < 0.05). Thus, non-parametric tests have been used for this study.

To answer the first research question (To what extent is the mental effort is impacted by AR?), an unpaired two-samples Wilcoxon test comparing the two groups’ mental effort scores was performed. To address the second research question (How does AR impact the students short-term and long-term learning outcome), an unpaired Wilcoxon test was performed to compare the learning outcome scores (post and retention) of both groups. To investigate the results and compare the evolution of scores in each group (experimental or control), paired samples Wilcoxon test have been performed. Furthermore, Spearman correlation tests were performed to analyze how mental effort can be related to learning outcome performances (research question 3: How does mental effort during learning tasks may influence scientific learning).

The significance level is considered at p = 0.05 or lower. More specifically, significant results have been interpreted as follows: p < 0.05, significant; p < 0.01, strongly significant; p < 0.001, very strongly significant; and p > 0.05, not significant. All of the statistical analyses were computed through the use of the R software V. 4.1.3 (R Core Team, 2020) using the Rcmdr package V. 2.8 (Fox and Bouchet-Valat, 2022).

Figure 6 illustrates that there were no significant differences in the mental effort scores between the experimental group and the control group. The average score for the experimental group is 4 (out of 1 to 7), SD = 1. The scores obtained by the control group is 4 (out of 1 to 7), SD = 1.2. There are no significant differences between the two groups [Wilcoxon test: T (16) = 62; p = 0.76].

As shown in Figure 7, the learning outcome scores increased for both groups from pre to short-term tests and from pre to long-term tests. The scores obtained for pre-test, short-term test and long-term test were, respectively: 2 (SD = 3), 8.5 (SD = 2), 4.4 (SD = 3) for the experimental group and 1 (SD = 4), 10 (SD = 3), 6 (SD = 5) for the control group. The paired Wilcoxon test underlines that the variations of both groups across the pre, short- and long-term tests are significant: first, experimental group (pre-test/short-term test: T (16) = 0, p = 0.0006; pre-test/long-term test: T (14) = 11.5, p = 0.01; short-term test/long-term test: T (16) = 2.5, p = 0.0007); second control group (pre-test/short-term test: T (17) = 0, p = 0.0003; pre-test/long-term test: T (17) = 4, p = 0.00059; short-term test/long-term test: T (16) = 6, p = 0.001). Not surprisingly, both groups’ modalities have a positive impact on short- and long-term learning outcome scores. Thus, test scores are significantly higher in short-term tests (experimental group = 8.5; control group = 10) than in the pre-tests (experimental group = 2; control group = 1) and long-term tests (experimental group = 4.4; control group = 6).

A comparison of the learning outcomes’ scores of both groups showed no significant difference for pre-and long-term scores (pre-test: H (1.33) = 1.4; p = 0.24; long-term test: H (1.33) = 2.18; p = 0.14, unpaired two-samples Wilcoxon). However, the control group significantly outperformed the experimental group in the short-term test score, showing a short-term negative effect of the AR on learning outcomes [H (1.33) = 5.64; p = 0.017]. This result suggests that RA can negatively impact knowledge in the short-term.

Figure 8 highlights that there was a negative correlation between the scores obtained on short-term test and long-term test, and the mental effort scores.

Although a moderate negative correlation between the pre-test scores and mental effort (r = −0.31; p = 0.077) has been detected, the correlation is not significant but close to the significance level of 0.05. The lack of correlation between short-term test and mental effort scores is clearer (r = −0.14; p = 0.44). A significant moderate negative correlation between mental effort scores and long-term learning outcomes’ scores is highlighted (r = −0.38; p = 0.031). Indeed, a link between mental effort and pre-test and shorts-term test is not underlined within the total sample. However, the correlation between mental effort and long-term test scores is significant. In other words, the long-term learning outcomes score is lower when the mental effort is higher, and vice versa.

The objective of this study was to investigate the impact of the AR on mental effort and learning outcomes scores in a learning context. Based on Sweller’s cognitive load theory Sweller (1988), we expected that AR would not increase the students’ mental effort and consequently impact negatively their learning outcomes in the short and long term. Indeed, because all of the students used this AR model for the first time, the risk of increasing mental effort was high. By analyzing the metal effort scores of both groups (i.e., experimental and control) we investigated the first research question (To what extent is the mental effort is impacted by AR during a science training learning?). Interestingly, our results based on the self-assessment metal effort test do not differ between the experimental group (using AR) and the control group. Even though the AR model was used for the first time, it was possible to drastically limit the mental effort (at least at the same level when not using AR models). In contrast to earlier findings (Van Bruggen et al., 2002; Hron and Friedrich, 2003; Sawicka, 2008; Wu et al., 2013), we find that using new learning material such as an AR model does not increase the on-task perceived mental effort during the investigation phase of the inquiry learning process.

The second research question (How does AR impact the students short-term and long-term learning outcome?) was investigated by analyzing the learning outcome scores of the experimental and control group. No significant difference was found by comparing the experimental group (with AR) and the control group (without AR) regarding long-term learning outcomes. Nevertheless, significant results were highlighted with better short-term learning outcome scores for the control group. Accordingly, it can be hypothesized that this lack of positive effect on learning can be attributed to a knowledge related loss of attention (Tang et al., 2003; Morrison et al., 2009; Radu, 2014). Our results demonstrate that the use of AR helps the students to focus on the pre-tests AR modeling task at the expense of the related content knowledge without impacting the perceived mental effort. Thus, we believe that a more specific guidance for the use of the AR model would be very helpful for the AR users by making the related knowledge more explicit, especially when they use the AR model for the first time. The construction of knowledge from learning activities without any guidance is possible, but the amount of guidance received has a direct influence on the student’s understanding of the concept that is addressed (Kirschner et al., 2006; Siméone et al., 2007; Moli et al., 2017).

We investigated the last research question (How does mental effort during learning tasks may influence scientific learning outcomes?) by investigating the correlation between learning outcome scores and mental effort score. Our findings revealed that the long-term learning outcome scores are low when the students’ mental effort is high (independently of group conditions, control or experimental), and vice versa. In other words, there is a link between learning mental effort and the long-term learning outcome score for the experimental and control groups. When the mental effort is higher, the long-term learning outcome score is lower. Conversely, when mental effort is lower, the long-term learning outcome score is higher. The correlation between pre-test learning outcomes and mental effort is close to significant. During the learning session activities, students who have better knowledge of the scientific subject are logically less cognitively loaded than the other students during the learning activities, and vice versa. No correlation was found between short-term test and mental effort, which shows that mental effort level preferentially impacted long-term outcomes. Our experiments are in line with previous results that suggest a correlation with long terms outcomes (Örün and Akbulut, 2019).

We are aware that our research may have three limitations. The first limitation is the restricted number of participants in this study, which would explain why some findings are unclear (type II error may occur). Non-parametric statistical tests limited our ability to detect significant differences. To support the results, it would be interesting to extend the sample size and perform parametric tests. The second limitation is the use of an indirect measure of mental effort by using a mental effort test during the training session. However, this way of assessing mental effort has already shown satisfying results (Paas and Van Merriënboer, 1994; Kalyuga et al., 2001). The duration of empirical study may be seen as short. However, this duration is the one usually devoted to study the phenomenon of the seasons with student teachers. Spending more time on seasonal phenomenon at the expense of other scientific topics is inappropriate considering the respect of the provided science teaching quality for the master students engaged in this study.

The third limitation is due to the instrument for assessing participants’ mental effort. It is clear that self-reported measures may be very different form direct measure in the sense of possible lack of correlation between perception of mental effort and mental effort (Yates et al., 2022).