Scene construction in healthy aging – Exploring the interplay between task complexity and oculomotor behaviour

Mounting evidence indicates a close correspondence between episodic memory, mental imagery, and oculomotor behaviour. It remains unclear, however, how oculomotor variables support endogenously driven forms of mental imagery and how this relationship changes across the adult lifespan. In this study we investigated age-related changes in oculomotor signatures during scene construction and explored how task complexity impacts these processes. Younger and cognitively healthy older participants completed a guided scene construction paradigm where scene complexity was manipulated according to the number of elements to be sequentially integrated. We recorded participants ’ eye movements and collected subjective ratings regarding their phenomenological experience. Overall, older adults rated their constructions as more vivid and more spatially integrated, while also generating more fixations and saccades relative to the younger group, specifically on control trials. Analyses of participants ’ total scan paths revealed that, in the early stages of scene construction, oculomotor behaviour changed as a function of task complexity within each group. Following the introduction of a second stimulus, older but not younger adults showed a significant decrease in the production of eye movements. Whether this shift in oculomotor behaviour serves a compensatory function to bolster task performance represents an important question for future research.


Introduction
Mounting evidence points to distinct differences in the way younger and older adults sample the visual world and subsequently deploy this information to support memory formation and retrieval (Açık, Sarwary, Schultze-Kraft, Onat, & König, 2010;Firestone, Turk-Browne, & Ryan, 2007;Wynn, Amer, & Schacter, 2020).For example, it is now wellestablished that older adults rely to a greater extent on prior knowledge or expectations to guide memory-related behaviours (Umanath & Marsh, 2014).During visual search tasks, this reliance on prior knowledge leads to preferential and repeated visual sampling of semantically congruent areas (e.g., searching for a toaster on the kitchen counter) but decreased memory accuracy and increased search time when items are displayed in semantically incongruent locations (e.g., a toaster displayed somewhere on the floor, Wynn, Ryan, & Moscovitch, 2020).Older adults are also more likely to direct their attention towards exogenously salient cues, even when explicitly instructed to ignore them or previously warned of their imminent appearance (Erel & Levy, 2016).Similarly, older adults have been found to produce more gaze fixations when exposed to novel stimuli during visual exploration, compared to younger adults (Liu, Shen, Olsen, and Ryan (2018).While some of these changes may reflect age-related deficits in cognitive function, evidence suggests that alterations in oculomotor behaviour sometimes serve a compensatory function to offset cognitive decline as we age and to potentially augment cognitive performance in response to increasing task demands (Wynn, Olsen, Binns, Buchsbaum, & Ryan, 2018;Wynn, Shen, & Ryan, 2019).This proposed mechanism may further support memory retrieval in cognitive aging by facilitating access to consolidated episodic and/or semantic representations, as indicated by studies on gaze reinstatement (Wynn et al., 2019).Accordingly, it has been observed that older adults tend to rehearse the eye movement patterns they produced when first exposed to a set of visual stimuli.Moreover, this behavioural response has been posited to assist with the successful recollection of such stimuli at later time points (Wynn et al., 2018).
To date, most eye tracking studies of memory-related processes in healthy aging have focused on experimental tasks involving the explicit presentation and subsequent recollection of visual stimuli.As such, it remains unclear how age-related changes in oculomotor behaviour manifest in the absence of visual cues, that is on tasks relying on the endogenous generation and maintenance of complex mental constructions.Scene construction refers to the capacity by which humans mentally construct and envisage richly detailed and spatially integrated three-dimensional representations (Hassabis & Maguire, 2007, 2009) and is suggested to underpin a range of complex cognitive endeavours, including episodic retrieval, future thinking, and spatial navigation (Mullally & Maguire, 2014).Given its complexity, the construction of three-dimensional spatially integrated scenes invariably draws upon several underlying cognitive processes, including the retrieval and reinstatement of episodic and semantic representations, their flexible recombination, as well as their active maintenance in the mind's eye (Ladyka-Wojcik, Liu, & Ryan, 2022;McCormick & Maguire, 2021).Functional neuroimaging and clinical lesion studies converge to reveal the importance of a distributed brain network, centred on the hippocampus, for the construction of spatially contiguous scenes (Hassabis, Kumaran, Vann, & Maguire, 2007;Irish et al., 2017;Mullally, Hassabis, & Maguire, 2012;Wilson et al., 2020).
Empirical research exploring age-related changes in scene construction capacity is scarce, although two studies warrant discussion in this context.First, Rendell et al. (2012) demonstrated that the overall capacity to construct atemporal scenes is significantly reduced in older compared to younger adults.The scenes generated by older adults contained significantly less contextual detail and were subjectively perceived as less salient and less vivid relative to their younger counterparts (Rendell et al., 2012).A separate study using visually presented verbal cues assessed age-related changes in relational processing during scene construction (Romero & Moscovitch, 2012).Results indicated that older adults generated fewer inter-item associations and displayed an increased propensity to omit item words when presented with larger item sets (i.e., six elements, as opposed to three, four or five).This reduced capacity to engage in multimodal integration was also associated with poorer performance on a subsequent cued recall test in the older group (Romero & Moscovitch, 2012).In view of these findings, the authors suggested that while constructive deficits in older adults largely reflect age-related changes in associative processing, the capacity to maintain a coherent mental representation of items in long-term memory may also moderate task performance (see also De Beni, Pazzaglia, & Gardini, 2007;Romero & Moscovitch, 2012).From a theoretical perspective, the ability to engage in mental constructive endeavours is further influenced by the accessibility and richness of conceptual knowledge (Irish, 2020;Irish & Piguet, 2013).It is now well established that such semantic representations assume an increasingly prominent role in the perception, encoding, and retrieval of external stimuli as we age (Umanath & Marsh, 2014), and likely provide the requisite scaffold upon which mental simulations are constructed (Irish & Piguet, 2013).
Collectively, the evidence to date points to age-related changes in the construction of mental scenes relying on internally generated imagery.However, to our knowledge, no study has explored the intersection between oculomotor behaviour and scene construction in healthy aging.This gap in the literature is surprising when we consider that vision is the primary modality by which scenes are realised.Indeed, the anatomical colocation of the visuo-oculomotor system and the core episodic memory system suggests considerable functional interplay between the reinstatement of visual imagery and the dynamic construction of events (Conti & Irish, 2021;Kinjo, Fooken, & Spering, 2020;Wynn, Van Genugten, Sheldon, & Schacter, 2022).Most importantly, recent neuroimaging studies in young adults indicate that unrestricted eye movements strengthen effective connectivity between hippocampal and oculomotor regions during scene construction, while constraining oculomotor behaviour significantly reduces the perceived vividness of the resulting scene imagery (Ladyka-Wojcik et al., 2022).The extent to which such dynamics shift throughout the aging process remains an open question.For example, age-related changes in the integrity of the episodic memory system have been shown to influence the efficiency of visual exploration patterns by reducing the amount and/or type of incoming information that is successfully captured and bound into a lasting memory representation (Ryan, Shen, & Liu, 2020;Ryan, Wynn, Shen, & Liu, 2022).In addition, studies on working memory demonstrate that the production of eye movements may interfere with mental imagery due to divided attention and competition for available cognitive resources (Andrade, Kavanagh, & Baddeley, 1997).As working memory capacities naturally decrease over the course of the lifespan (Daselaar, Dennis, & Cabeza, 2007), adaptive strategies favouring a reduction in visual explorations might underlie these changes.Accordingly, tasks with higher cognitive load may be predicted to elicit an attenuation of eye movement production.From a qualitative standpoint, seminal work from Spivey and colleagues suggests that, when listening to prerecorded instructions to imagine visual scenes in the absence of external visual stimuli, participants' eye movements tend to follow the directionality of the auditory description (Richardson & Spivey, 2000).Along the same lines, formal enquiries on what is now known as the "looking at nothing" effect strongly indicate that scan paths during visual imagery re-enact the original oculomotor patterns produced during perception (Laeng & Teodorescu, 2002).Taken together, these studies indicate that the (re-)instantiation of visual mental imagery relies on detailed, internal memory representations of spatial associations (Ferreira, Apel, & Henderson, 2008), although it remains unclear whether, and to what extent, this behavioural response is influenced by cognitive aging.
In the present study, we adapted a novel paradigm developed by Summerfield, Hassabis & Maguire (2010) to track how scene construction performance and concurrent eye movement behaviour vary as a function of task complexity in healthy older versus younger adults.Participants were required to mentally visualise and integrate an increasing number of auditorily presented elements into a spatially coherent scene while viewing a blank screen.We hypothesised that, irrespective of age group, when task demands are low, eye movements would support the visuospatial processing required to yield a coherent scene representation in the mind's eye.In contrast, when task demands increase above a certain threshold, we predicted a concomitant reduction in oculomotor behaviour, as indexed by a significant decrease in visual exploration.This prediction was motivated by our recent theoretical framework which proposes that on more cognitively demanding trials there would be a need to redirect available cognitive resources away from the production of eye movements to support the maintenance of a vivid scene representation in the mind's eye (Conti & Irish, 2021).Furthermore, we expected these changes to emerge specifically in the older group, reflecting a potential compensatory mechanism to support task performance in the face of declining cognitive resources.

Materials and methods
A cross-sectional study design was used whereby participants visited the FRONTIER clinic at a single point in time.Hsieh, Schubert, Hoon, Mioshi, & Hodges, 2013;So et al., 2018), a broad assessment of cognitive function covering domains of attention, memory, fluency, language, F. Conti et al. and visuospatial function.Exclusion criteria for both younger and older participants included concurrent psychiatric diagnosis, presence of other neurological syndrome, traumatic brain injury, or history of alcohol or substance abuse.All participants had normal or corrected-tonormal vision.

Ethics approval and informed consent statement
Participants provided written consent to take part in the study in accordance with the Declaration of Helsinki.The study was approved by the University of Sydney Human Research Ethics Committee (HREC) as part of the following two projects: "Memory and Imagination in Aging" (approval number 2018/479) and "Clinical Assessment for Aging and Neurodegeneration Research" (approval number 2020/224).

Apparatus
Participants were seated at a distance of 50 cm from a 1920 x 1080 resolution, 19″ Dell M991 monitor.The height of the chair and the chinrest used during the task were adjusted so that participants' gaze was directed to the centre of the screen as they sat in a neutral upright position.Monocular eye movements of the right eye were recorded using an Eyelink II eye tracking system (SR Research Ltd., Mississauga, Ontario), with a sampling rate of 500 Hz.Prior to the task, a 9-point calibration and validation procedure was conducted.If required, participants were allowed to wear glasses during the task provided no reflexes were detected during the calibration and the Eyelink could successfully recognise participants' pupils.The Eyelink built-in event detection feature was used to quantify participants' fixations and saccades.Fixation durations and saccade amplitudes were calculated based on the corresponding timestamps and X-Y coordinates of participants' gaze.

Task stimuli
In keeping with the Summerfield et al. (2010) study, stimuli comprised 360 audio recordings, divided into 180 construction elements and 180 control elements.Construction elements were short descriptions of objects commonly found in indoor settings.These could be pieces of furniture (e.g., "A blue sofa with cushions"; "A large round table "), everyday and/or decorative items ("A green leafy plant"; "A small grey radio"; "A pair of woolly gloves"), or background features ("A pink patterned wall"; "A cream stone floor").In contrast, control elements were short phrases whose combination of words was carefully designed to elicit minimal imagery or memory representations (e.g., "A single unit of quanta", "A principle of matter"; "An important type"; see Summerfield et al., 2010, for full task description).Given that the original audio stimuli were recorded by an English speaker for use in a UK population, we re-recorded the audio stimuli by an Australian-English speaker to avoid attentional disruptions due to unfamiliar pronunciations.
Regardless of the experimental condition (construction or control), each trial presented a minimum of 3, and up to a maximum of 6, audio stimuli, whose order of presentation was randomised across subjects.Participants performed a total of 80 trials (40 construction trials, 40 control trials), randomly intermixed and divided into 8 blocks of 10.Participants also conducted a practice test of 4 trials (2 construction trials and 2 control trials) to familiarise themselves with the task and, if needed, to request clarification on the experimental procedure.The stimuli presented in the practice test were not counterbalanced across subjects and were not used in the main experiment.

Procedure
Fig. 1 presents a schematic of the testing protocol used adapted from Summerfield, Hassabis & Maguire (2010).Testing proceeded largely in keeping with the original study design.Each trial began with the text "Clear your imagination" at the centre of the screen against a black background (2 s), prompting participants to prepare for the upcoming trial.This was followed by a cue indicating the current experimental condition, i.e., "Construct" for construction trials and "Attend" for control trials (1 s), and finally a fixation cross, displayed in the middle of the screen (1 s).At this point, the screen would turn black, and participants heard a series of audio stimuli, ranging from a minimum of 3 to a maximum of 6 elements.Each stimulus was followed by a short visualisation interval whose duration increased as the number of elements increased (i.e., 1 s after the first stimulus, 1.5 s after the second stimulus, 2 s after the third, fourth and fifth stimulus, and finally 2.5 s after the sixth stimulus).The duration of each visualisation interval was informed by the original Summerfield, Hassabis & Maguire (2010) experimental protocol.Importantly, participants did not know beforehand how many elements would be presented in any given trial.

Subjective ratings
Following the construction of each scene, participants were prompted to provide ratings about specific aspects of each trial using the keyboard.For construction trials, participants were asked to rate the vividness of the imagined scene (1 = not vivid,.., 5 = very vivid), the overall difficulty of the current trial (1 = easy,.., 5 = difficult), as well as the level of integration of the imagined scene (1 = not integrated,.., 5 = very integrated). 1For control trials, participants were asked whether a particular word was present or absent in the trial, e.g., "Was the word 'fragment' in the trial?"(1 = yes, 2 = no), and to rate the overall difficulty of the current trial (1 = easy,.., 5 = difficult).No time limit was imposed when answering these questions to accommodate slower search times in older participants.It is important to note that, unlike Summerfield, Hassabis & Maguire (2010), where participants performed the task during a functional MRI scan, our participants were instructed to keep their eyes open at all times, both for construction and control trials.They were allowed to move their eyes as desired, provided their gaze was directed towards the screen.Participants were only permitted to look away from the screen to locate the appropriate button on the keyboard when prompted to provide subjective ratings.

Post-experiment debriefing
Once the experimental task was completed, participants completed a short questionnaire to provide ratings (ranging from 1 to 5) on different aspects of the task.Questions were verbally formulated by the experimenter and digitally recorded for subsequent behavioural analyses.For construction trials, ratings were collected regarding the overall ease of visualisation and spatial integration of the different elements within the scene, the degree of familiarity, plausibility, coherence, and emotional salience of the imagined scene, as well as the viewer's perspective, the involvement of the self, and the level of adherence to the narrative description.For control trials, ratings included the imageability of phrases and the extent to which they elicited any memory representation.

Statistical analyses
Behavioural performance on the scene construction task was analysed using R (R Core Team, 2012, version 4.2.0) and MATLAB (The MathWorks Inc, 2017).Wilcoxon rank sum tests, as well as post-hoc tests 1 Participants were asked to provide subjective ratings on key phenomenological dimensions using a scale from 1 to 5. In keeping with the protocol implemented by Summerfield, Hassabis & Maguire (2010), only the upper and lower limits were explicitly stated (i.e., for vividness ratings the scale was displayed as "1 = not vivid, .., 5 = very vivid").
F. Conti et al. using the Benjamini-Hochberg procedure (Groppe, 2023), were performed to explore potential group differences between younger and older adults in terms of subjective ratings (i.e., vividness, difficulty, integration, and all behavioural variables included in the postexperiment questionnaire) due to the non-parametric nature of these data.Group differences on oculomotor measures of interest (i.e., number and duration of fixations, number of saccades and total scan paths) were established using linear mixed model analyses and Sidak post-hoc tests (R function afex, Singmann, Bolker, Westfall, Aust, & Ben-Shachar, 2022).For each oculomotor variable, we defined a separate linear mixed model with "Group" as fixed effect and "Subject Number", "Condition" (i.e., Construction or Control) and "Complexity" (i.e., the number of trial elements, ranging from 3 to 6), as random effects.Posthoc power calculations on sample size for eye-tracking data were conducted using G*power (Faul, Erdfelder, Lang, & Buchner, 2007).Results from these calculations revealed that a sample size of 32 participants was sufficient to detect a significant main effect of size f = 0.37, corresponding to η 2 p = 0.12, with 98% power and an alpha level of 0.05, as well as a significant interaction of size f = 0.40, corresponding to η 2 p = 0.14, with 99% power and an alpha level of 0.05.Finally, Spearman correlation analyses with Benjamini-Hochberg corrections for multiple comparisons were run to explore potential associations between oculomotor measures of interest and subjective ratings provided by participants.

Post-experiment behavioural results
Table 1 presents the subjective ratings provided by participants for construction and control trials on the post-experiment questionnaire.Overall, for construction trials, both participant groups found it relatively easy to clear their mind in preparation for the upcoming audio stimulus presentation.Once the narrative had begun, they tended to find it easy to imagine a single acontextual item, and moderately easy to integrate subsequent elements into the scene.Constructed scenes were typically imagined from a first-person visual perspective although participants did not subjectively feel immersed in the scene.Constructed scenes were rated by participants as emotionally neutral, primarily novel, highly plausible, and spatially coherent.Participants found it moderately easy to maintain all the presented elements within the scene and reported adhering to the task instructions by keeping to the narrative description.These patterns are in keeping with previous behavioural findings using the same task in healthy young participants (Summerfield et al., 2010).
Wilcoxon rank sum tests revealed no significant differences between younger and older adults across the post-experiment ratings (all p values > .05Benjamini-Hochberg adjusted).For control trials, the audio stimuli did not elicit rich mental imagery or detailed memory representations in either group.

Trial by trial scene construction performance
Table 2 displays the average vividness, integration, and difficulty ratings according to level of scene complexity for younger and older adults.Wilcoxon rank sum tests revealed that older participants subjectively rated their constructions as more vivid and more integrated than the younger group (all p-values < .01).Older adults also tended to rate construction trials as less difficult compared to younger adults (p = .07for trials containing 3 elements, then all p-values < .02for trials containing 4, 5 or 6 elements; see Table 2).Finally, Wilcoxon rank sum tests revealed that, irrespective of age group, self-reported vividness, and level of integration of the constructed scenes significantly decreased as the number of scene elements increased (both p-values < .001)whereas difficulty ratings significantly increased as a function of scene complexity (p < .001;see Supplementary Material for pairwise post-hoc comparisons based on the number of elements).

Control trial performance
For control trials, participants were prompted to pay attention and try to remember the stimuli that had been auditorily presented.Accordingly, trials with more elements placed greater demands on working memory.Wilcoxon rank sum tests did not reveal any significant group differences in terms of average performance accuracy across all control trials (3 elements: 70% Younger vs. 65% Older; 4 elements: 65.6% Younger vs. 63.1% Older; 5 elements: 68.7% Younger vs. 60.6%Older; 6 elements: 68.1% Younger vs. 59.4% Older; all p-values > .05).Similarly, a Wilcoxon rank sum test revealed no significant group differences in the perceived difficulty of control trials (p > .1,mean difficulty ratings for control trials provided in Supplementary Material).

Oculomotor analyses
We next analysed the eye movements produced in the gap intervals following each stimulus presentation.The oculomotor variables of interest were the number of fixations, the average fixation duration, the number of saccades and the total scan paths.In keeping with the experimental protocol of the original study, the duration of the gap intervals between stimuli increased as the number of elements increased to accommodate for the heightened working memory demands and overall complexity of the task.As a result, all oculomotor measures were calculated with respect to the duration of the corresponding trial.Furthermore, given that the distributions of the above-mentioned variables were skewed and heavy-tailed, we applied either a logarithmic (number of fixations, number of saccades and total scan paths) or an n th root transformation (fixation durations: n = 5) of the data prior to performing linear mixed model analyses.

Age-related differences in the number of fixations
A linear mixed model analysis revealed no significant main effect of Group (F(1,30) = 0.93; p = .34),Condition (F(1,30) = 2.04; p = .16),or Complexity (F(2.71,81.40)= 2.20; p = .10)in terms of the number of fixations participants produced (see Supplementary Table 3 for all model details and results).A significant Group x Condition interaction was found (F(1,30) = 4.88; p = .035),whereby older adults were found to produce more fixations in control versus construction trials compared to younger adults.

Age-related differences in the number of saccades
A linear mixed model analysis revealed no significant main effect of Group (F(1,30) = 0.97; p = .33),Condition (F(1,30) = 1.90; p = .18),or Complexity (F(2.72,81.56)= 1.78; p = .16)in terms of the number of saccades participants produced (see Supplementary Table 5 for all model details and results).Nonetheless, a significant Group x Condition interaction emerged (F(1,30) = 5.32; p = .028),whereby older adults produced more saccades in control versus construction trials compared to younger adults.

Table 1
Group averages in task ratings for construction and control trials provided by younger and older participants in the post-experiment questionnaire.Values represent group averages across all construction or control trials, with corresponding standard deviation provided in brackets.Ratings ranged from 1 to 5 and were coded as follows: difficulty: 1 = easy, 5 = difficult; perspective: 1 = 1st person, 5 = 3rd person; personal involvement: 1 = involved, 5 = absent; emotional salience: 1 = neutral, 5 = emotional; novelty: 1 = novel, 5 = familiar; coherence/plausibility: 1 = low, 5 = high; adherence to task narrative: 1 = kept to the narrative description, 5 = added new elements or context.For control trials, ratings also ranged from 1 to 5 and were coded for both mental imagery and memory representation as follows: 1 = low, 5 = high.Group differences were explored using Wilcoxon rank sum tests and Benjamini-Hochberg corrections.

Age-related differences in total scan path
We next calculated the total scan path participants produced in each trial as a function of trial duration.Scan paths were obtained by measuring the total distance the eyes covered in the allocated time, i.e., the sum of the amplitudes of participants' saccades.Given that, by design, trials comprising more elements took longer to complete, values were normalised by dividing each scan path by the duration of the corresponding trial.A linear mixed model analysis revealed no significant main effect of Group (F(1,30) = 0.35; p = .56,see Supplementary Table 6 for all model details and results).Nonetheless, a significant main effect of Condition was found (F(1,30) = 4.11; p = .05),whereby the amount of eye movements (AEM) participants produced was higher in control compared to construction trials (Difference between estimated marginal means in logarithmic scale: Construction -Control = -0.336).Furthermore, a significant main effect of Complexity was found (F (2.06,61.68)= 7.97; p < .001),indicating that the total AEM participants produced was higher in trials comprising 5 and 6 elements, compared to trials comprising 3 elements only (3 versus 5 elements: p = .01;3 versus 6 elements: p = .006;adjusted p-values > .1 for all other post-hoc pairwise comparisons).No significant interactions were found (all p-values > .2).
In summary, the linear mixed model analyses failed to reveal significant group differences in the total scan paths participants produced over the course of the experimental task, although they point to a significant increase in the AEM as the number of task stimuli increased from 3 to 6.Given that the auditory stimuli were presented in a sequential manner, it remains unclear whether the production of eye movements differed across groups in the early stages of the scene construction process (i.e., while participants were required to visualise and integrate the first 1 to 3 elements).
To determine whether this was the case and how the presentation of a new element affected participants' oculomotor response as the number of task stimuli gradually increased, we focused on pairwise differences between the scan paths participants produced in the gap intervals following the presentation of adjacent stimuli (see Fig. 2).We then built a separate linear mixed model for each variable.Given that the distributions of the resulting variables were skewed and heavy-tailed, we applied an n th root transformation of the data (n = 6) prior to performing linear mixed model analyses.
Interestingly, the linear mixed model analyses yielded significant results when comparing the difference in the AEM participants produced following the presentation of the first and the second stimulus (see Supplementary Table 7 for all model details and results).We found a significant main effect of Group (F(1,30) = 10.29,p = .003),whereby after the presentation of the second stimulus, the AEM decreased in older but not younger participants (Estimated marginal means of the difference in AEM between 1st and 2nd stimulus in the n th root scale, n = 6: Younger adults: − 0.58; Older adults: 13.94; Group difference: 14.5, p = .003).In contrast, no significant main effect of Condition was found (F(1,30) = 0.37, p = .55).Similarly, the Group x Condition interaction was not significant (F(1,30) = 0.93, p = .34).No significant main effects or interactions were found in terms of the difference in AEM past the presentation of the third stimulus (all p-values > .1).

Correlation analyses between subjective ratings and oculomotor measures
Spearman correlation analyses with Benjamini-Hochberg corrections for multiple comparisons were run to explore potential associations between behavioural and oculomotor measures of interest on construction trials for each level of complexity.Given that the linear mixed model analyses presented in the previous section failed to reveal a significant main effect of group, we aggregated data from our two participant groups prior to conducting the correlation analyses.
Our results revealed no significant correlations between the number of fixations and saccades participants produced or their total scan paths and the perceived vividness, integration, and difficulty of constructed scenes (all adjusted p-values > .05).On the other hand, in trials comprising 3 and 4 elements, the average fixation duration was found to be positively correlated with both the subjective vividness (3 elements: r = 0.14, p = .03;4 elements: r = 0.12, p = .05)and the subjective Fig. 2. Total scan paths produced by younger (left) and older (right) participants on construction trials following the sequential presentation of each auditory stimulus.Data are shown in the n th root scale (n = 6).Violin plots were obtained using the violinplot function for Matlab (Bechtold, 2016).
These results indicate that, on trials comprising fewer elements, maintaining fixation is associated with the elaboration and maintenance of the constructed scenarios in the mind's eye, thus enhancing the phenomenological experience of the mental endeavour.

Discussion
The objective of this study was to understand how oculomotor behaviour unfolds during scene construction performance in younger versus cognitively healthy older adults in the absence of externally cued visual stimuli, and to explore the impact of increasing task complexity on this relationship.Using an adapted version of the scene construction task developed by Summerfield, Hassabis & Maguire (2010), we demonstrated significant changes in patterns of visual exploration during the endogenous generation and maintenance of complex scene imagery.Notably, profiles of oculomotor behaviour in the early stages of scene construction differed in younger and older adults.We observed a significant reduction in the production of eye movements specifically in the older group immediately after the presentation of the second auditory stimulus, a behavioural response that may reflect an age-related recalibration of cognitive resources in response to task demands.Our findings resonate with recent theoretical frameworks proposing that visuo-oculomotor behaviour depends on available cognitive resources and is therefore highly susceptible to task influences in older age (reviewed by Conti & Irish, 2021;Ryan et al., 2022;Wynn et al., 2019).
The key finding from our study is that oculomotor behaviour during the endogenous construction of scene representations appears to be influenced by task complexity.Irrespective of age group, participants produced a greater amount of eye movements in control versus construction trials and as the number of task elements increased.Similarly, our analyses indicated that the durations of participants' fixations were longest in trials comprising a smaller number of elements (i.e., 3 or 4), and positively correlated with the subjective vividness and integration of the corresponding scenes.Longer fixations have been previously shown to index greater cognitive effort and focused attention (Dewhurst et al., 2018;Henderson & Hollingworth, 2003;Hollingworth, 2006).As such, we tentatively suggest that the paucity of cues provided during the initial phase of the scene construction process may require the assembly of an appropriate semantic scaffold, into which the elements can be integrated.During this process, maintaining fixation rather than engaging in active visual exploration may be beneficial, or indeed essential, to support task performance.As a result, fixations may be expected to decrease in number but simultaneously increase in duration.On the other hand, once a coherent spatial backdrop for the mental simulation has been established, the integration of an increasing number of elements into the scene can be more easily achieved.By this view, in the later stages of scene construction, more cognitive resources may be allocated to the production of eye movements, resulting in shorter fixations.Notably, eye-tracking studies on autobiographical memory retrieval have revealed that the recollection of emotional experiences tends to elicit more fixations and saccades, as well as greater imagery strength, compared to neutral events (El Haj, Nandrino, Antoine, Boucart, & Lenoble, 2017).Given that our task stimuli were largely divested of emotional and self-referential connotations (e.g., "A pair of reading glasses") the quantitative and qualitative differences in the production of eye movements we observe here are not surprising.While the present experimental task focused on scene construction reduced to its most foundational form, future studies exploring the impact of emotion and self-reference on these processes will be particularly revelatory.
Interestingly, looking at changes in participants' scan paths in the early phase of the scene construction process revealed a decrease in the production of eye movements in older adults immediately after the presentation of the second auditory stimulus.This attenuation of oculomotor behaviour may reflect the dynamic redeployment of cognitive resources as task demands exceed memory capacity to better support the construction and maintenance of complex scenes in the mind's eye (Engelhard, van den Hout, Janssen, & van der Beek, 2010).Accordingly, such a shift in oculomotor behaviour would be most relevant in situations where external visual cues are not available, as in the current task, thus placing increased demands on working memory capacity to maintain the scene representation.Notably, this age-related oculomotor response was not borne out on the phenomenological level as older adults perceived their mental constructions as more vivid and integrated than their younger counterparts, suggesting that the quality of the resultant scenes was not compromised.Studies incorporating eye tracking suggest that the vividness and emotionality of past and future mental construction is significantly reduced when guided eye movements are concomitantly executed (Engelhard et al., 2010;Kavanagh, Freese,Andrade,& May, 2001;Kemps & Tiggemann, 2007;van den Hout, Muris, Salemink, & Kindt, 2001), the argument being that this type of oculomotor activity selectively disrupts visuospatial working memory (Andrade et al., 1997).It is important to note that the findings reported here relate to instructed and tightly controlled forms of mental construction, and it remains unclear how spontaneous forms of oculomotor behaviour that are not specifically guided or cued might impinge upon the phenomenological experience.We propose that as the number of task elements increases, so too do the working memory demands of the task, reflecting the need to hold separate elements in mind, and to integrate these elements into a coherent and vivid mental representation.On such cognitively demanding trials, this recalibration of cognitive resources is posited to come at the expense of spontaneous oculomotor behaviour, leading to a decrease in accompanying eye movements.We note, however, that further studies formally testing the relationship between working memory capacity and oculomotor behaviour will be important to verify our hypothesis.Likewise, examining the impact of spontaneous versus instructed forms of oculomotor behaviour on participants' ratings of mental effort, as well as the accompanying subjective experience during mental construction will be an important future direction for this work.
Another counterintuitive finding of the present study is that older adults generated significantly more fixations and saccades relative to younger adults exclusively on control trials, where visual imagery generation was not explicitly cued.The production of eye movements has previously been shown to be guided by both visual and semantic representations (De Groot, Huettig, & Olivers, 2016).Importantly, activation of these representations occurs simultaneously when task stimuli are introduced verbally rather than visually (Huettig & McQueen, 2007), as is the case of the current study.It is therefore possible that the distinct profiles of oculomotor behaviour we observed on control trials reflect a general underlying tendency to prioritize either visual or conceptual representations during mental construction, a tendency that is likely to vary considerably across individuals and, potentially, age groups.It is now well-established that, as we age, domainspecific knowledge assumes an increasingly prominent role in driving how we perceive, process, and recall external stimuli (Umanath & Marsh, 2014).We speculate that this tendency to rely on semantic information to a greater extent may explain the propensity of older participants to generate more fixations and saccades than their younger counterparts on control compared to construction trials, where stimuli were designed to be abstract and more conceptual in nature (e.g., "A single unit of quanta", "A principle of matter").While these proposed mechanisms are speculative at this stage, our findings provide an initial hint that patterns of oculomotor behaviour might diverge in healthy aging in response to alterations in underlying cognitive mechanisms, particularly in the absence of externally cued visual stimuli.
Our findings stand somewhat in contrast to those of Rendell et al. (2012) in which older adults rated their imagined scenarios as less salient and spatially coherent than their younger counterparts.Task differences likely account for the above-mentioned discrepancies in subjective ratings, given that we used a more structured and carefully controlled experimental paradigm with well-defined cues to ensure participants adhered to the narrative description as much as possible (see also Sheldon & Levine, 2016).Whereas previous studies have interrogated the provision of contextual details as the main outcome measure on mental construction tasks, we constrained our focus to key oculomotor and phenomenological indices.It is therefore possible that, by its nature, our carefully controlled experimental design further mitigated age-related changes in associative binding and attention (Grilli & Sheldon, 2022;Irish, 2020).
Our findings add to a growing body of evidence suggesting that older adults leverage distinct profiles of oculomotor behaviour as a compensatory mechanism to support task performance (Ryan et al., 2022;Wynn et al., 2019).As previously mentioned, older adults are more likely to employ gaze reinstatement (i.e., rehearse the eye movement patterns they produced at an earlier time point, Wynn et al., 2019) to assist the consolidation and recollection of previously encountered visual information (Wynn et al., 2018;Wynn et al., 2019).Nonetheless, in the absence of externally cued visual stimuli, the findings of the present study point to distinct age-dependent oculomotor dynamics that may support mental construction.Specifically, when task demands increase beyond a critical threshold, it may become more efficient for older adults to conserve available cognitive resources by attenuating the production of eye movements and increasing efforts towards the construction and maintenance of a spatially coherent scene representation in working memory.Future studies will be required to delineate how scene construction dynamics unfold in the context of healthy aging and how this putative recalibration of resources influences the phenomenology of constructed scenes.Moreover, it will be essential to clarify the neural mechanisms driving the dynamic recalibration of resources in response to variations in task demands, and how such processes might be leveraged to predict and optimise cognitive outcomes in older age.
As the first study to investigate oculomotor behaviour during scene construction performance in healthy aging, several methodological considerations invariably warrant discussion.First, we acknowledge our relatively small sample size, which may have reduced study power to detect significant associations between the experimental variables under investigation.While previous studies exploring eye movements in healthy aging have included comparable sample sizes (i.e., 15-20 participants), we emphasise the need for future studies in larger cohorts to replicate our findings.Second, we did not include an objective index of the contextual details generated during scene construction.This was intentional as the focus of this study was to explore oculomotor dynamics as a function of task complexity using a carefully controlled paradigm.Third, we did not control for individual differences in the capacity for mental imagery in this study.This was intentional as by presenting participants with the precise contextual elements for integration into their mental simulations, we aimed to minimise potential confounding effects driven by individual differences in imagery capacity.We note that if the effects we observed in this study were driven purely by age-related differences in visual mental imagery, we would have expected the difficulty ratings provided by participants to reflect these deficits.Importantly, older adults rated their constructed scenarios as less difficult to elaborate compared to their younger counterparts.It is becoming increasingly clear that currently available verbal tasks assessing complex expressions of memory fail to capture adaptive neural and cognitive changes that naturally occur in older age, such as a shift from detail-to gist-based memory representations (see Andrews-Hanna, Grilli, & Irish, 2019;Grilli & Sheldon, 2022) suggesting that narrativebased tasks designed to elicit the provision of specific detail are not always suitable in the context of aging research (discussed by Miloyan, McFarlane, & Suddendorf, 2019).Nevertheless, it will be important to determine how objective and subjective aspects of scene construction relate to indices of oculomotor behaviour, and how such relationships change with increasing age and experience.Furthermore, we note that the stimuli used in the current study may be more familiar or meaningful to older, relative to younger, adults and may have enabled them to harness schemas or gist-based representations rather than constructing scenes de novo (Irish, 2020).The novelty ratings provided by participants at the end of the experimental task suggest that older adults generated scenes that were less novel compared to younger adults, which may also explain why they rated construction trials as less difficult and experienced their imagined scenarios as more vivid and integrated.As a result, it may be that the richer repository of semantic and general conceptual world knowledge of older participants affords the necessary scaffold to generate familiar scenarios, in line with the semantic scaffolding hypothesis (Irish, 2016;Irish & Piguet, 2013).Experimental tasks requiring the extraction and flexible integration of multiple sensory and visuospatial elements reveal significant impairments in older adults (Castellano, Guarnera, & Di Nuovo, 2015), as do tasks requiring the recombination of episodic details into novel scenarios (Addis, Musicaro, Pan, & Schacter, 2010).It remains unclear how older adults might fare on experimental tasks that disproportionately tax the construction of novel or implausible atemporal scenarios and we suggest this will be an exciting area for future studies (discussed by Renoult, Irish, Moscovitch, & Rugg, 2019).Moreover, given that our experimental paradigm focused exclusively on the endogenous generation of visual mental imagery, future studies directly comparing endogenously generated versus externally cued visual imagery will be crucial.Finally, experimental paradigms employing a different setup which allow for greater freedom of movement (as opposed to a mounted chinrest) may provide valuable insights regarding additional oculomotor variables of interest.Relevant variables include the spatial distribution of eye movements produced by participants and how they relate to the subjective experience of the mental simulation.

Conclusion
This is the first study, to our knowledge to explore the intersection between age-related changes in scene construction and visuooculomotor behaviour in the context of internally generated mental imagery.Our findings suggest a compensatory shift away from the endogenous production of eye movements, especially in older adults, as a function of increasing task complexity.Future research is required to determine the underlying neurocognitive mechanisms driving this agerelated reallocation of resources during complex feats of mental construction.
F. Conti et al.

Fig. 1 .
Fig. 1.Task design.Representative timeline of a construction trial in which four elements must be integrated into a coherent scene representation.Adapted from Summerfield, Hassabis & Maguire (2010).

Table 2
Scene construction ratings by scene complexity for younger and older participants.
a Ratings based on 40 total construction trials per participant, with 10 trials per level of complexity.Subjective ratings for construction trials were provided on a scale from 1 (lowest score) to 5 (highest score).Group differences were determined using Wilcoxon rank sum tests (W).Original p-values and p-values following Benjamini-Hochberg corrections are shown, where * p < .05;** p < .01;*** p < .001.

Table 3
Correlation analyses between average fixation duration and subjective ratings on construction trials.