What makes a visual scene more memorable? A rapid serial visual presentation (RSVP) study with dynamic visual scenes

ABSTRACT The visual system has been characterized as having limited processing capacity. Research suggests that not all visual information is equal and that certain visual scenes are registered better than others. The present study investigated how people process biological stimuli under time constraint using a Rapid Serial Visual Presentation (RSVP) paradigm with dynamic movie scenes. The results of Experiments 1 and 2 indicated that recognition memory as well as identification performance got better with longer duration (400 ms vs. 200 ms). Most importantly, biological stimuli led to better memory and lower reaction times. Lastly, Experiment 3 was conducted to replicate previously observed dynamic advantage and to disentangle the role of motion from content. The results indicated that dynamic scenes were remembered better than static scenes for both the biological and non-biological stimuli. The ecological validity and retrieval benefit of dynamic scenes were discussed in relation to image memorability.

Cognitive mechanisms that underlie fast processing of visual information depend on many factors.Those are extensively studied in the literature to understand the limits of the human visual system (Awh et al., 2006;Greene & Oliva, 2009;Kastner & Ungerleider, 2000;Potter, 1975;1976;Potter et al., 2004).The current research explores how the human visual system perceives moving stimuli using dynamic scenes extracted from movies.Due to their inherent complex form and similarities to everyday visual perception, studying dynamic visual scenes provide a valuable tool to gain better insight into the early visual perception.As humans make 3-4 saccades every second, this requires the visual system to integrate those separate fixations into a visual continuum (Henderson et al., 1999;Henderson & Hollingworth, 2003).Sequences of brief dynamic stimuli lasting a few hundred milliseconds have analogous properties to how the visual system deals with this integration.Also, limited research exists on how dynamic scenes are remembered until time constraints and early research suggests a dynamic advantage (Candan et al., 2015;Matthews et al., 2007Matthews et al., , 2010)).Literature on image memorability suggests that intrinsic qualities of visual information lead to changes in memory performance (Asp et al., 2021;Bainbridge et al., 2013;Kolisnyk et al., 2021;2023;Parikh et al., 2012;Rust & Mehrpour, 2020).In relation, characteristics of dynamic stimuli require more inquiry to further understand what makes dynamic images more memorable.Accordingly, the present research focuses on memory for dynamic stimuli involving biological and non-biological qualities under very fast presentation rates using a Rapid Serial Visual Presentation (RSVP) paradigm.

Memory for visual scenes
Early research on visual scene perception shed light into how people categorize and register visual information at fast exposure rates (DeLucia & Maldia, 2006;Goldzieher et al., 2017;Greene & Oliva, 2009;Gronau & Shachar, 2015;Intraub et al., 1992;Potter et al., 2002).Studies showed that people are good at identifying the presence of an object category when they are exposed to a natural scene very briefly, almost at a glance (Fabre-Thorpe, 2011).In addition, the schema of a scene also supports the observer's interpretation and perception according to the constructivist approach to visual perception (Intraub & Berkowits, 1996).Thus, it was suggested that people combine information based on mental schemas as well as fixations while gathering information about a visual scene.Research showed that spatial and global features of a scene are processed earlier than local features (Oliva & Torralba, 2001;Torralba & Oliva, 2003;Torralba et al., 2006).In a related work, Greene and Oliva (2009) explored the early visual processing of glances at natural scenes.They claimed that visual scenes as a whole can also be processed as units and recognized as a category as fast as individual objects.The researchers asked participants to categorize target images based on their basic-level or global properties.The results of this study indicated that global properties are perceived earlier than basic-level identification.In other words, when we look at natural scenes such as a lake, forest for a short time, we first process global features such as transience, navigability, openness and temperature; however more time is needed to get details like a forest scene versus a mountain scene.This suggests that early conceptual activation is an important factor in fast categorization of visual scenes.
Recent research goes a step forward to use deep neural network models to identify how neural models can learn from visual stimuli and recognize and identify visual information (Oliva & Schyns, 2000).This has both behavioural and neurological implications for perception of visual scenes.There is also a growing literature on the differences in memorability of visual stimuli (Asp et al., 2021;Bainbridge et al., 2013;Bylinskii et al., 2015;Isola et al., 2011;Kolisnyk et al., 2021Kolisnyk et al., , 2023;;Parikh et al., 2012;Rust & Mehrpour, 2020).Research using electrophysiological measures showed that frontal theta amplitude and occipital alpha amplitude were affected by individual-specific encoding success, while frontal positivity was affected by stimulus-intrinsic memorability and individual-specific encoding success (Kolisnyk et al., 2021(Kolisnyk et al., , 2023)).New research from Bylinskii et al. (2015) attempted to compute image memorability from visual data compared to human data.Also, research into face perception showed that distinctiveness is an important feature that contributes to the memorability of faces and that memorability of face photographs should not be reduced to specific pictorial and personality characteristics but should be treated as qualitatively different (Bainbridge et al., 2013).There has also been research on visual working memory and factors that affect the limit and capacity depending on visual information characteristics (Alvarez & Cavanagh, 2004;Asp et al., 2021).A recent study by Asp et al. (2021) showed that when visual complexity was controlled, people's visual working memory capacity expanded for meaningful stimuli (i.e., faces).Using an EEG measure, authors have observed that memory performance has increased and there was higher activity for meaningful stimuli contrasting faces and meaningless shapes.This suggested that distinct features of visual stimuli manipulate visual working memory capacity and that all visual stimuli are not equally processed.

Rapid serial visual presentation
The limits of visual scene perception are commonly studied with the paradigm called Rapid Serial Visual Presentation (RSVP) (Meng & Potter, 2008;Potter et al., 2002;Potter, 2012;Wyble et al., 2011).A series of items (words, letters, photographs etc.) are presented sequentially in an RSVP at fast exposure rates (50-500 ms).When visual items are presented at faster rates, subjects experience interference coming from the previously presented stimuli, called conceptual masking.Previous research has consistently found almost perfect memory for visual scenes when they were presented for 1 s (Potter et al., 2004;Standing, 1973).However, at shorter presentation times, ranging from 100 to 300 ms for each item in a series, recognition can be quite poor (Potter & Levy, 1969;Potter, 1975;Potter et al., 2002).Performance only improves when the presentation of stimuli is around 400 ms each (Potter & Levy, 1969;Potter et al., 2004).The researchers argued that although the processing of all the details of the visual scene needs more time, the gist of the scene can be extracted very rapidly.Potter (1976) argued that 300 ms is needed to successfully consolidate visual scenes into memory and avoid conceptual masking.A store called Conceptual Short-Term Memory (CSTM) was proposed, which relies on top-down influence and is dependent on the context, familiarity, and meaning of visual stimuli (Potter, 1976;2012) 300 ms was identified as an optimal duration for memory consolidation and conceptual processing.CSTM works by activating the memory of related concepts.This is consistent with the idea that we use prior knowledge to compensate for the impoverished visual information that we face every day.Subsequent research showed that contextual information helps compensate for the lack of visual information, as in the case of occlusion (Meng & Potter, 2008).This is also apparent in how subjects falsely recognized conceptually similar items as old, which again stressed the role of gist information that is extracted very early (Potter et al., 2004).Consistently, in a rather recent study, Gronau and Shachar (2015) found that recognition memory for very briefly presented objects (24 ms) was better for the contextually related objects than the contextually unrelated objects.This again supports the fast extraction of conceptual information when processing glimpsed visual stimuli.
Another important finding of this literature was the difference between recognition and identification performance with respect to the ease of the task.Potter (1975) showed that at 125 ms presentation rate, more than 70% of the target pictures were correctly identified when subjects were given a title of a picture (e.g., baby sitting on a chair) beforehand as opposed to recognition at that rate, which was at 13%.This difference has long intrigued researchers.What is the nature of the information that gets activated early in the visual system?Research favoured the idea that identification of a specific object affects detection and engagement of attentional resources (Potter et al., 2010).Potter and Hagmann (2015) showed that even at 13 ms/target, the detection of a target picture was above chance performance when a basic level category label instead of superordinate level was given prior to an RSVP.This further indicates that conceptual activation at the basic level category facilitates early detection.Potter and Fox (2009) also observed global activation of more than one simultaneous item in an RSVP which later led to secondary serial processing.In a more recent study, Goldzieher et al. (2017), who investigated repetition blindness (RB) for visual scenes in the first 100-150 ms, observed that scenes were encoded with local visual details rather than global category specific details.
Research provides mixed results for what happens in the early visual perception in terms of the activation of contextual information.In relation, we investigated how the nature of a moving stimulus, namely the motion category, can affect what gets registered under time constraints.

Memory for dynamic scenes
What happens when the image is not static but a dynamic one?In his ecological approach, Gibson (1979) stressed the relation between an agent and its environment with respect to how each object affords possibilities for action.We interact with the environment through active movement.The question remains about how this relates to the perception of dynamic stimuli in the early visual system.Few studies investigated the relative importance of dynamic images for memory (Candan et al., 2015;Goldstein et al., 1982;Kwok & Macaluso, 2015;Matthews et al., 2007), while most of the literature focused on the perception of static images and used well-defined photographs (i.e., people, indoor and outdoor scenes, animals etc.) (Meng & Potter, 2008;Potter et al., 2004;Potter & Hagmann, 2015).In contrast to early expectations, dynamic images were in fact remembered better than static ones.As dynamic images contain more information which may provide a burden on cognitive resources, they may also be processed in a qualitatively different manner that activates conceptual memory.Moreover, motion attracts attention (Abrams & Christ, 2003) and is characterized as having ecological validity (Gibson, 1979).
In their pioneering study, Goldstein et al. (1982) compared long-term recognition memory for moving and static pictures.The researchers expected to find better memory for static images because they provide less information as opposed to dynamic images, which contain more information, and might be harder to encode.Contrary to their expectations, they found that recognition memory was the best for dynamic stimuli.In a related study, Furman et al. (2007) showed that people remembered most events from audio-visual film narratives from 3h to 9 months interval, showing that long-term memory was robust memory for dynamic scenes.Consistently, Matthews et al. (2007) found that recognition memory was better for moving images compared to static ones at one week retention intervals.The researchers called this dynamic superiority effect.They argued that one reason why moving images might be remembered better was because they provide different angles as it relates to viewpoint invariance or that the moving stimuli attract more attention, therefore leading to better encoding.To test for these explanations, Matthews et al. (2007) used multistatic stimuli that provide implied motion instead of actual motion.As also demonstrated by Lander and Bruce (2003), continuous motion was remembered better than multistatic images shown without interstimulus interval (ISI).This led the researchers to conclude that fluid motion that has spatiotemporal characteristics was what led to better memory due to the activation of motion schemata.This suggests that motion provides qualitatively different information than the sum of static images.A related phenomenon called study-test congruence effect suggests that using the same type of stimulus at retention and test leads to better memory.Mathews et al. (2010) demonstrated that dynamic superiority effect and study-test congruence effect were unaffected by divided attention at encoding and that motion led to better spatial orientation judgments as well as memory for people and objects in the dynamic scene.
Motion information has been treated as ecologically valid and relevant according to vision researchers.As studies suggest a memory advantage for dynamic stimuli, the question becomes what are the characteristics of those stimuli which make them more memorable.So, we can be more receptive to action sequences as they can be more meaningful, conceptually rich, and ecologically valid.This has implications especially concerning image memorability research.We can suggest that the dynamic nature of a visual stimulus can also be considered a unique characteristic that could alter memory capacity based on memorability.Among dynamic stimuli, biological quality of dynamic images can play a role in increasing memory performance.

Biological motion perception
Biological motion is a special type of motion that is ecologically significant and conceptually important.This type of motion is attributed to living organisms which move in relation to their environment (Bardi et al., 2011;Johansson, 1973;Troje & Westhoff, 2006).Biological motion has been studied intensely in the literature by using behavioural methods that utilized point-light walkers, which showed that people are perceptive to biological motion (Bertenthal & Pinto, 1994;Blake & Shiffrar, 2007;Bromfield & Gold, 2017;Casile & Giese, 2005;Johansson, 1973Johansson, , 1976;;Troje, 2013).Johansson (1973) defended that human motion patterns could be analyzed with points of light positioned at joints without any contribution from pictorial information.Gender (Barclay et al., 1978;Mather & Murdoch, 1994) as well as emotion (Atkinson et al., 2007) among others can be identified rather easily from these types of point-light walker displays.Casile and Giese (2005) observed that recognition of biological motion can be accomplished by relying on coarse spatial details.Additionally, studies using physiological methods such as TMS (Transcranial Magnetic Stimulation) and fMRI (Functional Magnetic Resonance Imaging) also showed that people are sensitive to biological motion and differentiate it from other sources of motion, and brain has specialized parts like STS (Superior Temporal Sulcus) as well as FFA (Fusiform Face Area) (Grossman & Blake, 2001;Grossman et al., 2005), which specialize in biological motion.Perception of biological motion also benefits from experience.Studies of Cutting et al. (1988), and Hiris and Cramer (2005) have shown that naive participants could not recognize biological motion by looking at single static frames while the non-naive ones performed significantly better in the same task.Biological motion also degrades with age similar to other cognitive faculties.In a recent study, Agnew et al. (2020) investigated the relationship between age, attentional abilities, and biological motion perception using various test batteries.Although older adults were slower than younger adults, this decline in performance was not observed for accuracy.According to the correlation analysis, there was no statistically significant correlation regarding accuracy and reaction times between visual attention and biological motion perception tasks.In other words, the decline in biological motion perception with age is not explained by the attention factor.
Among other aspects, form might have an effect in perception of biological motion.To study the role of underlying form on motion perception, Hiris (2007) used two different types of biological motion displays: translational and treadmill motion and four different non-biological displays: unstructured translation, unstructured rotation, structured translation, and structured rotation.Results have shown that adding form to non-biological motion leads to better performance compared to unstructured non-biological motion.Even if structured translation performance was not as good as translating biological motion, it supported the idea that adding form to non-biological motion increases detection performance, making it more similar to biological motion.
What role does biological motion play in memory for briefly presented naturalistic dynamic scenes?Troje (2013) states that using point-light displays has an advantage since they isolate the kinematics of a movement from the other sources of information available in more real-life like displays.So, research from point-light displays may not be replicated with naturalistic scenes.Research is lacking and controversial with respect to whether biological motion contributes to short-term memory for dynamic images.As biological motion is suggested to have an underlying coherent form (Hiris, 2007), it may be crucial in activating what Matthews et al. (2007) call motion schemata, therefore leading to faster and better encoding and recognition.Consequently, the current study focused on the conceptual identification of dynamic images and explored whether biological motion provides a conceptual cue for rapid categorization and recognition of visual scenes.

Experimental overview
In a series of three experiments, we investigated recognition and identification of dynamic images displaying biological versus non-biological motion.We explored the role of biological motion as a factor increasing memorability under time constraints.We operationalized biological motion as any action attributed to animate agents as opposed to non-biological motion as any action attributed to inanimate objects.Animacy is generally defined as the distinction between living and non-living entities.In these experiment series, we employed movie scenes as naturalistic stimuli that involve rich visual characteristics (Cutting et al., 2012;Cutting, 2022).As the research, which employs movie stimuli to understand the underlying cognitive mechanisms of attention, spatial cognition and event perception is growing (Bordwell & Thompson, 2003;Levin & Baker, 2017;Smith et al., 2012), interest in movies as a type of visual media (comics, movies, video games, etc.) take up interest due to their similarities to real-life action.Although sequences from movies are framed and may not be considered as naturalistic as daily actions, they provide valuable comparison to everyday motion perception and can tell us about how we process dynamic images in the early visual system.
The study motivated to be a follow-up to a previous experiment conducted by Candan et al. (2015), where researchers questioned whether dynamic images coming from movie sequences lead to better memory compared to their static counterparts.That paper showed that dynamic images were recognized better than static images, especially at 400 ms presentation interval, and that memory performance declined with faster presentation times.This suggests that motion helps recognition of visual scenes when resources are limited.Also, it further proposes that there may be a critical interval (∼400 ms) for conceptual short-term memory for dynamic images.We can argue that motion sequences have simply more information and may provide redundant cues.Dynamic scenes are more naturalistic and have richer information in comparison to static frames.They attract attention due to motion onset (Abrams & Christ, 2003) and facilitate encoding due to contextual facilitation (Biederman, 1972;Chun & Jiang, 1998 ; Smith & Vela, 2001;Smith & Manzano, 2010;Smith et al., 2014).Not only can we remember them better, but they may be critical time intervals where we can process motion information in a meaningful fashion.As suggested by Potter (1976), a minimum of 300 ms critical interval may be needed to process information in a conceptual manner.This also may be due to a critical interval in which one can complete a recognizable action that unravels in time.But are all motions equal?We may be more receptive to motion cues because of their ecological significance and thus humans may recognize and identify biological motion more easily compared to non-biological motion.The visual system processes different types of motion in visual scenes such as motion derived from biological entities (i.e., man kicking a ball) as well as inanimate objects (i.e., waves crushing on a shore).Because animacy has special ecological value for humans (Bardi et al., 2011;Casile & Giese, 2005;Johansson, 1973;Troje, 2013), we can suggest that motion attached to biological entities would have priority over inanimate objects.
In Experiment 1, we used two presentation speeds (200 vs. 400 ms) and contrasted recognition memory for biological versus non-biological motion in coloured as well as black and white movie scenes.In Experiment 2, we provided a verbal cue before the RSVP stream and investigated identification performance for the same stimuli.As suggested by the literature, the recognition of biological motion is done with ease by humans (Blake & Shiffrar, 2007;Bromfield & Gold, 2017;Hiris, 2007).We hypothesized that biological motion would be recognized and identified better and faster than non-biological motion when resources are limited.Consistent with early literature, we also hypothesized that stimuli presented for 400 ms should be remembered and identified better than briefer stimuli presented for 200 ms.We also expected that recognition performance would be poorer than identification performance, which activates relevant conceptual information, and requires the monitoring of a single stimulus.
In both experiments, we also contrasted the role of colour in recognition and identification performance with the hypotheses that coloured dynamic scenes would be remembered and identified better than black and white stimuli both for biological and nonbiological motion.Literature indicates that diagnostic colour affects scene recognition (Oliva & Schyns, 2000).In relation, Wichmann et al. (2002) showed that subjects were better at recognizing coloured images than black and white images.Similarly, Spence et al. (2006) observed that when presented from 20 ms to 2000 ms, people detected coloured neutral images better than black and white neutral images.In movies, multiple perceptual information is presented simultaneously as characteristics of shot lengths and camera motion; colour evolved over the years (Cutting et al., 2012).Although the role of colour in visual scenes was researched more extensively (Oliva & Schyns, 2000;Wichmann et al., 2002), how colour relates to the memorability of dynamic scenes is an open question.Certain studies showed that colour of movies affect cognitive mechanisms such as event segmentation (Cutting et al., 2012).As early era movies were mainly black and white, the technology introducing colour contributed to bringing movies closer to what we perceive in the natural world and improved aesthetics and appeal of these types of media (Cutting et al., 2012;Cutting, 2021Cutting, , 2022)).As coloured scenes may provide more retrieval cues, they can facilitate the conceptual identification of the nature of motion.
Lastly, Experiment 3 served two purposes.The first one was to replicate previous research (Candan et al., 2015), which compared recognition memory for dynamic versus static scenes.Consistent with early findings, we expected that dynamic stimuli would be recognized better than static stimuli.Secondly, we intended to disentangle motion information from the characteristics of biological and non-biological information.To achieve that we have compared the static and dynamic versions of both types of stimuli to get a more in depth understanding about whether the action properties were the underlying factor behind the memory advantage observed in Experiments 1 and 2. In this experiment, we explored the role of biological "motion" over non-biological "motion" in dynamic and static stimuli.We operationalized a static frame taken from a dynamic scene as implied motion.The properties of a static picture can cause it to be perceived as having potential for motion.To illustrate with an example, a static picture of a woman on stairs can have potential for movement up or down the stairs.In a recent work, Khatin-Zadeh (2021) showed that static pictures containing cues specific to a particular movement evoked more motion perception (e.g., a picture of a man while stretching before running, the perception of movement when a driver sees a photo of a car standing on the road).Implied motion perception was also activated when the participants saw a canvas painting with brushstrokes where a drawn picture was associated with movement.It was found that those can activate the motor areas of the brain (Thakral et al., 2012).Therefore, we hypothesized implied biological motion would be recognized better than implied non-biological motion.

Transparency and openness
We reported our design, outlier detection, data analysis, and participant samples in the following experiments.The stimuli, data, scripts and data analyses were made available in the OSF link: https://osf.io/pj2uv/.All experiments were approved by the Ethics Committee of Yaşar University, and participants provided their informed constants for each experiment.

Participants
The number of participants necessary to detect an effect size of f = .25 with an alpha level of .05 and a power of .80 was calculated using G*Power (Faul et al., 2007) and was found to be 24.In total, 81 Psychology students from Yaşar University completed Experiment 1 (65 female; 16 male; age M = 21 years, SD = 2.1).All participants had normal to correctedto-normal vision and none of the participants were colour blind.Outlier analysis led to two of the participants being excluded from the analyses due to having higher z scores than 3.29 in any of the measures taken (Tabachnick & Fidell, 2013).

Design
Experiment 1 had 2X2X2 mixed design with duration (200 ms vs. 400 ms) treated as a between-subjects, colour (coloured vs. greyscale), and motion (biological vs. non-biological) treated as within-subjects variables.Sensitivity (d ′ or d prime) and reaction time (in milliseconds) were measured.

Materials
The materials were part of a large database previously compiled by Cutting et al. (2010) and later used by Cutting et al. (2012) and Candan et al. (2015).A total of 400 clips (200 tests and 200 distractors) were chosen from 20 movies, half of which showed biological motion while the other half showed non-biological motion.Biological stimuli were chosen as examples of motion of a living organism (i.e., person walking, person jumping, animal running etc.) while non-biological motion was chosen as examples of motion of objects, which were conceptually identifiable (i.e., car passing, waves crushing, plane flying etc.).Half of the clips were in black and white, and the other half were in colour (See Figure 1 for stimuli examples).For the duration condition, 200 ms stimuli consisted of 5 frames presented in succession while 400 ms stimuli consisted of 10 successive frames.All clips had 256 × 256-pixel size with 24 frames/second presentation rate.The clips were constructed to come from the middle of a given shot with either 200 ms or 400 ms duration, which led to 800 clips in total for both duration conditions.10 research assistants, blind to the purpose of the study, coded the clips for the validity of the manipulation.

Procedure
Matlab Psychophysics Toolbox was used to programme and conduct the experiment (Brainard, 1997).The stimuli were shown on a HP computer screen running on windows 10 operating system with 21.5 inch (54,61 cm).The screen had a resolution of 1920 × 1080 pixels, with a refresh rate of 60 Hz.The experiment took place at Psychology Department Laboratory of Yaşar University.After coming to the laboratory and being greeted, participants were asked to report their age, gender, visual acuity and colour blindness.
The stimuli were presented using an RSVP (Rapid Serial Visual Presentation) procedure without any interruption.Subjects have performed 40 trial blocks, and, in each block, they were presented with either 5 biological or 5 non-biological motion clips in a random order.Order of movies was randomized for each participant.Each trial block started with a 200 ms black fixation cross for the participant to attend to the middle of the screen and the background colour during the whole experiment was grey.The fixation cross was followed by a 200 ms blank screen before the RSVP stream and the presentation ended with a white noise mask (See Figure 2 for an example trial block).Immediately after each presentation, the subjects performed a recognition test where they saw each clip separately.As found by Matthews et al. (2007), we have used matching presentation and test stimuli in the form of moving images to benefit from study-test congruence effect.Before the test items, the participants were presented with the following question "Have you seen the following clip?"After each clip, participants responded "Yes" or "No" to 10 items, 5 tests (coming from the previous presentation) and 5 distractors (coming from the same movie) using the right and left arrow keys.Reaction times (in milliseconds) were also recorded for each test trial and those were collected from the moment participants saw the answer choices on the screen until they pressed the button to respond.There was a 500 ms blank screen between each trial block.Before starting the experiment, participants partook in a practice block to familiarize themselves with the study procedure.A coloured biological motion presentation was chosen for the practice and after each test item, 2000 ms-long feedback was provided on the screen.Participants were free to repeat the practice block or pass to the experiment phase.
Reaction times.There was a statistically significant main effect of motion (F (1, 79) = 4.63, p = .035,η 2 = .055)and duration (F (1, 79) = 63.47,p < .001,η 2 = .445).Participants were faster for clips showing biological motion compared to clips showing non-biological motion.Also, reaction time for 200 ms clips were higher compared to 400 ms clips.No significant effects for colour or any of the interactions were observed (See Figure 4).

Discussion
The results of Experiment 1 showed that dynamic clips involving biological motion were both recognized easier and faster than dynamic clips involving non-biological motion for both durations.This observation suggests that people might be more sensitive to registering biological motion when resources are limited.This result should be interpreted with caution though.As the stimuli in our biological motion condition are complex, they include other features besides motion per se for their memorability.First, they include animate agents, mainly humans, as the subject of the related action.The effect can therefore also be attributable to the presence of human agents as well as the ease in which people have been known to identify and remember faces (Bruce & Young, 1986;Grill-Spector et al., 2004;Hay et al., 1991;Haxby et al., 2000;Yovel & Kanwisher, 2004).Therefore, to suggest that it is the biological motion that facilitates memory under time constraints, we need to disentangle the effect of motion from the other characteristics of these biological stimuli, which we intended to do in Experiment 3.
The results of Experiment 1 also indicated that recognition performance has declined with faster presentation duration as participants performed better with 400 ms stimuli for both stimuli type.This result is consistent with earlier studies using RSVP for image recognition (Meng & Potter, 2008;Potter et al., 2002;Potter, 2012;Wyble et al., 2011).Also, we observed that reaction times have increased for faster stimuli.The motivation behind collecting reaction time data was to see whether there was a potential speed-accuracy tradeoff concerning the two types of stimuli.We can say that as memory performance and reaction times indicate better accuracy in relation to faster reaction times, we cannot talk about a speed accuracy tradeoff for our stimuli.This shows that people were faster to react to biological stimuli, possibly due to the significance of these stimuli for the visual system.As reaction time measures can also allude to the certainty of the participants, further confidence measures are needed to make this judgment.In addition, we also found that colour is a factor when people process dynamic scenes.Coloured dynamic scenes were recognized better than black and white scenes for both stimulus types, consistent with colour advantage observed for visual scenes (Oliva & Schyns, 2000;Wichmann et al., 2002).
Overall, we can say that a critical exposure window (around 400 ms) is present for better encoding and retrieval of biological motion as well as non-biological motion.Although, biological motion still promotes better memory over non-biological motion even for faster presentation times as 200 ms.In Experiment 2 we questioned whether the memory facilitation observed in Experiment 1 expands to easier tasks such as identification, which helps activate the conceptual knowledge early by the presentation of a verbal description.

Experiment 2
The motivation behind Experiment 2 was to investigate the role of biological motion, exposure duration and colour on identification performance.As the literature suggests, identification of a specific scene described beforehand with a conceptual description is both easier and faster compared to recognition (Potter & Levy, 1969;Potter, 1975;Potter et al., 2004).Research is still lacking on understanding how the nature of the visual scene affects the identification performance and whether biological motion would provide a beneficial role on identification.

Participants
The number of participants necessary to detect an effect size of f = .25 with an alpha level of .05 and a power of .80 was calculated using G*Power (Faul et al., 2007) and was found to be 24.A total of 85 Psychology students from Yaşar University participated in Experiment 2 (75 female; 10 male; age M = 21.98 years, SD = 1.23).All participants had normal to corrected-to-normal vision and none of the participants were colour blind.Outlier analysis led to five of the participants who had z scores higher than 3.29 in dprime scores (Tabachnick & Fidell, 2013) being excluded from the study.

Design
A 2X2X2 mixed design was used in Experiment 2. Duration (200 and 400 ms) was manipulated as a between subjects factor and colour (coloured and greyscale), motion (biological and non-biological) were treated as within-subjects factors.Sensitivity (d ′ or d prime) and reaction time (in milliseconds) were measured.

Materials
The materials were the same as the Experiment 1. Differently, for each trial block (biological vs. nonbiological), a unique verbal description, in the form of a few words, (i.e., "people dancing," "balls spinning") which corresponds to each clip (target present trials) was created as well as distractor descriptions that do not correspond to any of the 5 clips in the presentation (target absent trials) (See Figure 5 for an example trial block).

Procedure
The procedure of Experiment 2 was similar to Experiment 1 with the following exceptions.Participants performed a total of 80 trial blocks, 40 were target present and 40 were target absent blocks.Each presentation again contained 5 clips in each of the six conditions.In each block, a verbal description that either belonged or did not belong to a single target was presented for 5000 ms before the presentation of the RSVP stream.Participants' task was to indicate whether the described action was either present or absent in the presentation by pressing the right and left arrow keys to indicate "YES" or "NO" respectively after each presentation.Reaction time measures (in milliseconds) were also taken for each response.The order of the conditions as well as the location of the target within the presentation were randomized for each participant.

Results
Sensitivity.The main effects of motion (F (1, 83) = 11.28,p = .001,η 2 p = .12),duration (F (1, 83) = 27.64,p < .001,η 2 p = .2) and the interaction of motion and colour (F (1, 83) = 37.28, p < .001,η 2 p = .31)were statistically significant.Identification of biological motion was better compared to non-biological motion and longer clips (400 ms) were identified better compared to shorter clips (200 ms).Participants had higher sensitivity to biological motion when it was presented in black and white instead of in colour (See Figure 6).Reaction times.There was a statistically significant main effect of motion for reaction time (F (1, 83) = 27.79,p < .001,η 2 p = .25).None of the other main effects and interactions were significant The results indicated that reaction times for non-biological motion was higher compared to biological motion regardless of colour and duration (Figure 6(d)).

Discussion
Consistent with the literature, identification performance was superior to the recognition performance on all measures.This is reflected in the ceiling effect observed for dprime values.The results of Experiment 2 showed that biological motion is identified better and faster than non-biological motion at both duration times.This indicates that when a pre-given verbal description is introduced before an RSVP stream, biological motion was identified better.This can be due to multiple reasons.As biological and non-biological stimuli both contain conceptual information, we can suggest that the information value of biological stimuli may be higher compared to nonbiological stimuli.This may again be related to the evolutionary value of quick identification of biological motion for survival purposes (Johansson, 1973).Although we again observed an advantage for stimuli type, the effect size was much smaller compared to Experiment 1.This is notable because it may be linked to potential mechanisms behind the superiority of biological information.As Experiment 1 used a recognition memory task, this difference may be rooted in memory compared to detection in Experiment 2, which is based on attention.Even though the effect of biological information is still there for attention as well, we can suggest that the type of information benefits memory more.
In addition, while identification of 400 ms-long stimuli were better compared to 200 ms-long stimuli for both motion types, we did not observe facilitation of presentation time on reaction times.This may be due to the relative ease of the task compared to recognition.More importantly, the memory benefit of biological motion was still apparent for 200 ms duration.Interestingly, we observed an interaction between colour and motion.This may be due to already high performance for the coloured stimuli where the nature of the motion might have helped less compared to when the stimuli were in black and white.As Experiment 1 and 2 both used dynamic scenes, we further investigated the role of biological actions for static scenes to see if the results replicate for implied motion, in the form of static images that have potential for motion.

Experiment 3
This experiment had two aims.One was to replicate the previous study by Candan et al. (2015) which investigated the role of motion in short-term memory using an RSVP.Second was to disentangle the effect of motion from the other characteristics of the biological stimuli, to be able to say that it is really the biological motion that facilitates memory under time constraints.In this experiment, we contrasted recognition memory for static and dynamic scenes involving biological and non-biological stimuli.In this experiment, the term "motion" for the static stimuli refers to implied motion.As static scenes only depict one instance of an otherwise dynamic scene, it has the potential to imply the conceptual nature of the potential action.Thus, this experiment investigated whether the implication of biological motion has superiority over nonbiological implied motion for recognition performance.

Participants
The number of participants necessary to detect an effect size of f = .25 with an alpha level of .05 and a power of .80 was calculated using G*Power (Faul et al., 2007) and was found to be 24.A total of 47 undergraduate Psychology students from Yaşar University participated in the Experiment 3 (43 female; 4 male; age M = 21.74 years, SD = 1.64).All participants had normal to corrected-to-normal vision and none of the participants were colour blind.According to outlier analysis, none of the participants were excluded from the experiment.

Design
Two-way repeated measures analysis of variance was conducted in the Experiment 3. A within-subjects factor of action (two levels: dynamic and static) and a within-subjects factor of motion (two levels: biological and non-biological) were analyzed as independent variables.Sensitivity (d ′ or d prime) and reaction time in milliseconds were recorded.

Materials
The materials were taken from the same database as Experiment 1 and Experiment 2 with the following exceptions.All dynamic clips featured coloured stimuli.The colour movies from Experiment 1 and Experiment 2 were used with the addition of 10 new movies, for a total of 20 movies.All the stimuli in this experiment were presented for 400 ms.The reason for the narrowing down of the stimuli was due to the established superiority of coloured stimuli and longer presentation durations.For the static condition, the first frame of each clip was used and remained on the screen for 400 ms.

Procedure
The procedure was similar to Experiment 1 except that in the static condition, all 5 stimuli were still images.The order of the static and dynamic conditions was randomized for each participant.For the practice block, biological motion stimuli consisted of dynamic clips only, while non-biological motion stimuli were all static frames.
Reaction times.The main effect of action for reaction times was statistically significant (F (1, 46) = 52.66,p < .001,η 2 p = .534)but no significant main effect was found for motion (F (1, 46) = .13,p = .72,η 2 p = .003)(Figure 7(d)).Further, the interaction between action and motion was not statistically significant (F (1, 46) = .88,p = .35,η 2 p = .019).Therefore, participants took more time to respond to static stimuli compared to dynamic stimuli but there were no differences in reaction times for responding to biological over non-biological motion stimuli.

Discussion
Experiment 3 replicated the results of Candan et al. (2015) by showing that recognition performance was higher for dynamic stimuli compared to static stimuli.Also, biological stimuli were recognized better than no-biological stimuli.In addition, we observed a marginally significant interaction between action and motion.This showed that dynamic scenes were remembered better than static scenes for both biological and non-biological stimuli, but the effect size was bigger for biological stimuli.This suggests that even if an image depicts implied motion, visible motion is still needed for memory to benefit from biological as well as non-biological information.The more pronounced effect in the case of biological motion can be attributed to predictions concerning the conceptual identification of the nature of the movement.When we see a static image of a person (i.e., raising a hand), it may be difficult to predict what the nature of the movement would be as there are many possible options.The static view of a non-biological entity (i.e., waves, wheel) on the other hand may be more informative as to the possibilities of the ensuing movement.We should be careful to not over interpret this result though as the interaction was marginally significant and would require further replication.Lastly, regardless of the motion type, mean reaction times for still images were higher than dynamic stimuli.In other words, participants were faster in reacting to dynamic images regardless of whether biological and non-biological motion is depicted.

General discussion
We conducted a series of three experiments to investigate how people perceive and remember biological motion under time constraints.Results of Experiment 1 showed that biological motion was recognized better and responded to faster than nonbiological motion for both 200 and 400 ms presentation durations.This observation suggests that people might be more sensitive to registering biological stimuli when resources are limited.We should exercise caution when interpreting these results.As biological stimuli in our study also include other crucial futures such as the presence of humans and faces, therefore the effect can also be attributable to the presence of human agents as well as the ease in which people have been known to identify and remember faces (Bruce & Young, 1986;Ellis, 1975;Farah et al., 1998;Hay et al., 1991).Early work has shown consistently that face perception is special, and that people are prone to easily recognize identity, emotional state, speech, focus of attention (Bruce & Young, 1986;Ellis, 1975;Farah et al., 1998).The special nature of face perception is supported by physiological evidence such as a dedicated neural area, the FFA (Fusiform Face Area) as well as behavioural evidence such as face inversion effect (de Gelder & Rouw, 2000;Grill-Spector et al., 2004;Haxby et al., 2000;Kanwisher, 2000;Perrett et al., 1985;Rossion et al., 2012;Yovel & Kanwisher, 2004).Recent evidence also supports that people can interpret emotions from facial expressions (Krumhuber et al., 2023) and there are individual differences when it comes to face perception (Flessert et al., 2023;Murty et al., 2020;White & Burton, 2022).Therefore, the nature of the stimuli besides the motion type might be a contributing factor for the ease of recognition.
We also observed a duration effect as slower stimuli were recognized better than faster stimuli regardless of motion type.We can say that although biological motion information led to better memory, it still had constraints.More exposure time is needed for better memory just as in the case of non-biological motion.Experiment 1 also indicated that colour was a factor as coloured scenes resulted in higher memory performance compared to black and white scenes.This result is in line with the literature that has found colour was crucial in when people need to identify a scene category at fast presentation rates (Oliva & Schyns, 2000;Spence et al., 2006;Wichmann et al., 2002).
Moreover, Experiment 2 inquired whether the same results also extend to identification, which is a rather easier task.First, the performance was better in the identification task compared to the recognition task as it is apparent in the d prime scores.This is in line with the previous literature, which observed that the identification task was much easier than the recognition task (Meng & Potter, 2008;Potter et al., 2002).More importantly, the results indicated that biological motion was identified better and responded to faster than non-biological motion at both presentation durations and the duration effect was again significant.For both stimulus types, slower stimuli (400 ms) were identified better than faster (200 ms) stimuli.Interestingly, while lower in performance, identification of biological motion at 200 ms interval also resulted in higher than chance performance.That signals that while we can talk about a critical interval of around 400 ms for best recognition and identification performance as suggested by CSTM store (Potter, 1976(Potter, , 2012)), biological motion benefits memory even at faster presentation rates as low as 200 ms.One can argue that as people interact with the environment through motion as the ecological theory of Gibson (1979) suggests, we may have learned very early to identify actions in others.Motion perception also has an evolutionary advantage for survival where time becomes of the essence when you process actions.In addition, we have also observed an interaction between colour and motion as biological motion only helped identification when the stimuli were black and white.This may be due to the relative ease of the identification task where the performance was close to the ceiling and that biological motion only benefits identification when the stimuli are harder to identify due to deprivation of colour.
Lastly, our aims were two-fold for running Experiment 3. The first aim was to replicate the results obtained by Candan et al. (2015), which contrasted dynamic stimuli with static stimuli using an RSVP and observed a dynamic superiority.The second was to examine whether motion affects recognition for biological and non-biological stimuli.In this experiment, when we employ the term biological motion in static stimuli, we refer to implied biological motion.In Experiment 3, we replicated earlier findings and observed that dynamic scenes were remembered better than static scenes.We know that motion attracts attention (Abrams & Christ, 2003), potentially leading to better encoding.This may be one factor why dynamic stimuli are processed better than static stimuli.The results are consistent with the dynamic superiority effect proposed by Matthews et al. (2007).As research using RSVP with visual scenes has concentrated on well framed static photographs, movie stimuli are more complex, less framed and definitive.So, movie scenes may share more similarities with how we encounter motion information in everyday life.Referring to the image memorability research (Asp et al., 2021;Bainbridge et al., 2013;Kolisnyk et al., 2021;2023;Parikh et al., 2012;Rust & Mehrpour, 2020), the dynamic nature of visual stimuli appears to be a crucial feature, which increases the potential for these stimuli to be registered and later recognized.This can be stated as one of the important findings of this research.
We also observed a marginally significant interaction between action and motion.This suggested that, even though dynamic versions were remembered better than static versions for both biological and non-biological stimuli, the difference is more pronounced for biological stimuli.We should be careful not to over interpret this result though as the effect is weak and it needs replication in a further study.We can however suggest that human agents provide more degrees of freedom into the possible action one takes, so dynamic scenes might be more informative than static scenes in the case of biological stimuli.Also, as static scenes involving biological agents also have properties such as posture and bodily cues, the improvement in memory may be attributable to the animacy of the stimuli as dynamic clips evolve over time and may present a better representation of the underlying action (i.e., man walking).As Experiment 1 could not pinpoint the reason for the biological advantage, Experiment 3 indicates that dynamic scenes involving such agents are better remembered compared to static counterparts.This suggests that visible motion could have added another cue for later recognition.
This observation adds to our understanding of the limits of visual processing and the kind of visual information that are recognized in early visual stages.This raises the question of why dynamic stimuli are better remembered compared to static stimuli.This may relate to the added information provided by action for the otherwise uncertain nature of the movement for biological entities.As we are exposed to moving organisms in everyday life, visible motion may be a necessary factor for the biological nature of the stimuli to benefit memory.Literature also suggests that humans identify biological motion faster in point-light displays (Bardi et al., 2011;Casile & Giese, 2005;Johansson, 1973), our research shows that they also remember this type of motion faster and with ease when the nature of the stimuli are dynamic.This suggests that humans are sensitive to dynamic cues that may provide faster and better conceptual identification.

Limitations and future directions
One limitation of the current study relates to the nature of the stimuli.Biological motion is more common and varied in movie scenes as opposed to non-biological motion, which is more limited.As we have varied the selection by using a large stimuliset, a more controlled manipulation by creating videos in a laboratory environment or using computer animations can answer those questions.Also, visual stimuli coming from movies in general are different: they are more complex and less refined, unlike wellstructured photographs of visual scenes.They are also more dependent on what is going on in the rest of the scene.Previous research tells us that clutter affects recognition in movie scenes (Cutting & Armstrong, 2016).Researching how people differentiate biological motion from the background can provide insight into the attentional resources in the face of dynamic stimuli.One other limitation can be attributed to the choice of stimuli, which were subjectively coded by independent coders.More research is needed to identify a more mathematical operationalization of motion rate such as the visual activity index used by previous research (Cutting et al., 2012).This can give us more information about how much motion is required for people to conceptually recognize and identify an action in order to set a threshold for recognition.Lastly, as both of our dynamic stimuli were conceptually identifiable motion, the present research still leaves one question open.Does the benefit gained by dynamic images relate to mere motion or the conceptual value of the information?Future research can explore memory for motion without any conceptual value to better understand the role of motion in short-term memory.

Conclusions
Perception of dynamic images provides parallels to the complexity of information we face every day.Therefore, how well and how quickly we remember naturalistic dynamic images requires further investigation.The present research examined how people recognize and identify moving images under time constraints.Of those, the nature of the information (biological vs. non-biological) was investigated for its memorability.We showed that the biological nature of stimuli provided a better cue for recognition of visual stimuli at fast presentation rates as low as 200 ms.This was possibly due to the characteristics of human agents such as facial and bodily cues.The results replicated the dynamic superiority effect, extending it to both biological and non-biological stimuli.This indicates that dynamic nature can be a more potent cue compared to the nature of the visual stimuli.This motivates future research that can answer questions about the image memorability criteria.

Figure 1 .
Figure 1.Stimuli example for biological and non-biological motion.(a) black and white biological motion (b) coloured biological motion (c) black and white non-biological motion (d) coloured non-biological motion

Figure 3 .
Figure 3. (a) Mean sensitivity for black and coloured clips.(b) Mean sensitivity for clips depicting biological and non-biological motion.(c) Mean sensitivity for clips presented 200 and 400 milliseconds.

Figure 4 .
Figure 4. (a) Mean reaction time for clips depicting biological and non-biological motion.(b) Mean reaction time for clips presented for 200 and 400 milliseconds.

Figure 5 .
Figure 5.An example trial block for the identification task.

Figure 6 .
Figure 6.(a) Mean sensitivity for clips depicting biological and non-biological motion.(b) Mean sensitivity for 200 and 400 milliseconds clips.(c) Mean sensitivity for black and coloured clips depicting biological and non-biological motion.(d) Mean reaction time for clips depicting biological and non-biological motion.

Figure 7 .
Figure 7. (a) Mean sensitivity for clips presenting dynamic and static action.(b) Mean sensitivity for clips depicting biological and nonbiological motion.(c) Mean sensitivity for clips presenting biological and non-biological motion with dynamic and static action.(d) Mean reaction time for clips depicting dynamic and static action.