Background

Alzheimer’s disease (AD), the most common cause of dementia, and mild cognitive impairment (MCI), its preceding stage, are characterized by progressive decline in cognitive function. Although the most commonly reported symptom in AD is episodic memory deficits, objective impairment can be observed in several linguistic and cognitive domains during the course of AD and MCI, including but not limited to attentional control, semantic memory, and executive functions (Perry & Hodges, 1999; Verma & Howard, 2012). While memory complaints have been well documented, other areas of cognition have received relatively less attention. Neuropsychological tests that are currently in use measure attention, episodic memory, language, but employ artificial, laboratory tasks (e.g., word list learning and object naming). These typically do not target combinatorial or inferencing abilities. Moreover, existing tests are devoid of any context and do not reflect naturally occurring experiences or situations. Activities of daily living are an important component of diagnosing AD and determining the ability of individuals to function independently (Lawton & Brody, 1969). Current assessments of functional ability comprise of self-report or caregiver report. More objective assessments, that identify the root of the deficits and can point to appropriate interventions, are warranted.

Previous research has shown that individuals with AD have difficulty integrating smaller units of information into a whole. This has been observed in visual and spatial perception (Duffy, Cushman, & Kavcic, 2004; Paxton et al., 2007). However, identifying a whole given smaller units plays a role in many other cognitive domains, as for example in discourse comprehension. On a more abstract level, the ability to construe larger conceptual units is fundamental to all kinds of prediction phenomena (Kuperberg, 2021).

Event recognition and prediction are closely related to the ability to perform actions (Bailey, Kurby, Kurby, Giovannetti, & Zacks, 2013; Cooper, 2021). In order to perform an activity, one needs to have a relevant mental representation of the activity, often referred to as event model, which is activated based on an abstraction of the common features of past experiences called event schema (Zacks & Tversky, 2001). These schemata are created by drawing upon pre-existing experiences from episodic memory and semantic knowledge (Gerwien & von Stutterheim, 2018; Sargent et al., 2013; Zacks, Speer, Swallow, Braver, & Reynolds, 2007) and provide a framework for processing and encoding ongoing events (Zacks et al., 2007). Further, events can be segmented into smaller event units. Every event can in principle be a macro-event for a number of micro-events which are causally and temporally related to each other (Kuperberg, 2021). From this perspective, event knowledge is organized hierarchically. Making connections between micro-events and predicting subsequent events are key components involved in holistic processing of information. This is accomplished via inference generation, referred to as bridging inferences and predictive inferences, respectively (Cohn, 2019; Magliano, Dijkstra, & Zwaan, 1996; Singer & Halldorson, 1996).

What is the role of language in activating event schemata? Previous research has shown that if participants are prevented from using inner speech while performing cognitive tasks that involve – among others – information categorization, action planning, or switching between different tasks – performance typically goes down (Alderson-Day & Fernyhough, 2015). This points to an involvement of language. However, it is not yet entirely clear whether it is language use per se that supports performance because it helps to focus attention on relevant aspects of the task (Miyake, Emerson, Padilla, & Ahn, 2004), or whether it is really the specific semantic content that is transported via language, which supports performance. Since verbal tests have been used to assess cognitive decline previously, and some correlation between verbal behaviour and cognitive decline has been attested, one aspect of the current study will be about the role of language in event schema activation as measured in a non-verbal sorting task with and without verbal cueing.

Event cognition in ageing and Alzheimer’s disease

Studies investigating the effects of ageing and age-related neurodegenerative diseases on event cognition have been rather limited. One area of focus has been event segmentation ability and subsequent memory for these events (Bailey, Zacks, et al., 2013; Kurby & Zacks, 2018; Zacks, Speer, Vettel, & Jacoby, 2006). The evidence from these studies suggests that cognitively healthy older adults maintain their event segmentation abilities when compared to AD patients. In these studies, participants were presented with 1-min films and asked to segment them into smaller, meaningful events. Individuals with AD segmented the stimuli in a more idiosyncratic way, that is, agreement about where event boundaries should lie was low. This was in contrast to the younger adults group, who had very high segmentation agreement, as well as the older adults group, who also had higher segmentation agreement compared with the AD group, but less than the younger adults. Moreover, several studies found that individuals’ later memory was highly correlated with their segmentation ability (Flores, Bailey, Eisenberg, & Zacks, 2017; Kurby & Zacks, 2018; Zacks et al., 2006). These findings indicate that memory problems in AD occur not only due to issues related with storage and retrieval, but they actually begin with failure to encode information correctly.

Event cognition in AD patients has also been studied using a paradigm involving verbal scripts. The tasks involved script generation, wherein participants produced the steps in a given macro-event (Grafman et al., 1991; Roll, Giovannetti, Libon, & Eppig, 2019), or verbal script sequencing, in which participants either arranged the steps in event scripts in the correct sequence or determined whether the provided script sequences were correct (Allain et al., 2008; Grafman et al., 1991). The results showed that AD patients’ performance was worse on script generation and sequencing compared with cognitively healthy older adults. One recent study further eliminated the verbal aspect and replaced the verbal scripts with pictures depicting a sequence of actions in an event, which were to be arranged in the appropriate order (Roll et al., 2019). The findings remained consistent with verbal sequencing studies.

Studies using such tasks go beyond traditional episodic memory measures and psychometric tests (Sargent et al., 2013; Zacks et al., 2006) due to the richer and more realistic context they provide. In the current study, we wanted to explore further aspects of event cognition in AD patients, and additionally in MCI patients. Instead of segmenting a whole into smaller units, we focussed on the integration of smaller units into a whole (West & Holcomb, 2002). Using a paradigm similar to the sequencing paradigm in the Roll et al. study, we were interested in how AD and MCI patients in comparison with healthy older adults can activate and use schema knowledge to identify macro-events from their sub-component micro-events, and use schema knowledge for establishing temporal and causal sequences. Event segmentation and event integration, both, require knowledge of and the ability to activate event schemata. However, during segmentation, the cognitive system is alert to changes in the perceptual input in order to establish event boundaries. Integration, on the contrary, requires abstraction over several event units that are each separated by an event boundary. Segmentation and integration are linked via the theoretical concept of ‘granularity’. For example, if we consider the event of writing a paper, depending on different levels of granularity, either boundaries are established between hitting the letter ‘t’, hitting the letter ‘y’ and so on, or between typing and looking up references, or between working on a paper and taking a coffee break. Event integration is a cognitive activity that comprises of switching to a coarser granularity level, which allows integrating micro-events into macro-events by activating appropriate event schemas.

We conducted two experiments using picture narratives that examined participants’ ability to integrate parts of the events, draw upon event schema, and constitute the parts into their whole (Experiment 1); and, with a cognitively less demanding paradigm (Experiment 2). In both experiments, we studied patients diagnosed with AD, patients diagnosed with MCI, and cognitively healthy older adults.

Experiment 1

The aim of Experiment 1 was to investigate AD and MCI patients' ability to integrate micro-events to form macro-events and compare their performance with cognitively healthy older adults (OA group). A second goal was to examine whether cueing the appropriate macro-event schema via language improved performance. We developed a novel procedure in which participants were presented with four pictures that depicted sub-events representing different temporal stages within a single, larger macro-event, presented in a scrambled order. We assessed participants’ sequencing and naming accuracy.

Method Participants

A total of 47 participants completed the study. The cognitively healthy older adults were recruited from a database at the Network Aging Research, Heidelberg University. Potentially eligible participants were contacted via telephone, and a brief telephone screening was conducted to rule out visual or auditory impairment, any existing diagnoses of neurological or psychological conditions (e.g., stroke, depression, epilepsy, and Parkinson’s disease), or self-reported cognitive complaints. Eligible individuals were then invited to take part in the study. Before beginning the experiment, participants in the cognitively healthy group were administered the Mini–Mental State Examination (MMSE) to screen for possible cognitive impairment. Individuals who scored ≤26 points on the MMSE were later excluded from the analysis. A total of 25 cognitively healthy older adults completed the study. However, data from only 20 participants were included in the analyses, as three participants were excluded based on their MMSE score, and a further two participants’ testing session was interrupted due to a technical error.

The AD and MCI patients were recruited from the memory clinics of University Hospital Heidelberg and Central Institute of Mental Health, Mannheim. A total of 22 people were recruited, of which 10 had a diagnosis of AD, 10 were diagnosed with MCI, and two were excluded due to an uncertain diagnosis. The patients were formally diagnosed by a physician following neurological and neuropsychological evaluations. These included blood tests, clinical history, CT/MRI scans, cerebrospinal fluid testing, and CERAD test battery (Morris et al., 1989). The National Institute on Aging and the Alzheimer’s Association (NIA-AA) criteria were applied for AD diagnosis (McKhann et al., 2011). All individuals received a diagnosis of probable AD. Only individuals in the mild stage of AD (MMSE score: >19) were included in the study. The individuals in the MCI group were diagnosed according to the NIA-AA criteria for MCI (Albert et al., 2011) or the ICD-10 criteria (World Health Organization, 1992). The two criteria are comparable enough for participants to be grouped together. The classification of sub-type of MCI was only available for three of the ten participants, all of whom had amnestic MCI. For demographic characteristics, refer to Table 1.

All participants were native speakers of German and reported normal or corrected-to-normal vision. The group with AD had significantly fewer years of education compared with the cognitively healthy group, but MCI group’s education did not differ from either group. All participants provided informed consent. The study received ethics approval from the Ethics Commission of the Medical Faculty of Heidelberg University, Germany.

Stimuli

The stimuli consisted of 14 sets of events, each comprising four pictures depicting sequential stages within a macro-event. For example, in a trial depicting a ‘grocery shopping’ event, the pictures illustrate preparation of a shopping list, going to the supermarket, picking groceries, and paying. Each picture was 300 pixels in width and height. For example stimuli, see Figure 1. The stimuli were created from step-by-step tutorials demonstrating how to perform different activities, which were obtained from ‘WikiHow’ (http://wikihow.com). Initially, 40 such trials were created, which were tested in two groups – university students and university professors. During these test sessions, participants were presented the individual pictures of a set in a scrambled order and were asked to reorder them to resemble a temporally correct sequence, followed by naming the macro-event being depicted. The 28 events on which participants performed best in terms of naming agreement were selected – 14 in Experiment 1 and 14 in Experiment 2 (see below). The temporally and causally appropriate sequences were determined by the responses that maximum number of participants agreed upon during the pilot test. All, but one, events included in the final experiment had a sequence agreement of at least 95%, and one had an agreement of 86%. Macro-event names for verbal cueing were also determined by the results from the pilot test (for a full list, see Supplementary Materials).

Example of stimuli (a) scrambled presentation, (b) correct sequence.

On half of the trials in the actual experiment with patients and controls, participants were given the macro-event name as a cue, and only had to reorder the pictures. On the other half, they had to name the event after reordering the pictures. For the purposes of counterbalancing, two versions of the experiment were created. In one version, one half of the trials were presented along with the word cue, and the other half had to be named, whereas this was reversed on the second version. Each participant performed only one of the two versions, but these two versions were counterbalanced among the participants.

Procedure

Participants provided informed consent for the study, following which they filled out a sociodemographic questionnaire. In the cognitively healthy older adults group, this was followed by administration of the MMSE, and finally, the main experimental task was administered.

The experiment was programmed using JavaScript and presented on a 10.1″ screen tablet. At the start of the experiment, participants received detailed instructions for the task on the screen and additionally were given an explanation verbally when the task was not clear. They were given five practice trials at the beginning to ensure that they understood the task, were able to perform it correctly, and to familiarize them to using the tablet. Instructions were repeated as many times as required. Only after ensuring that participants understood the task, the actual experiment began.

Each participant was presented with fourteen trials – seven cued and seven non-cued – in an order that was randomized for each participant. The presented picture order within each trial was also randomized for each participant. The participants sorted the pictures using the touchscreen feature of the tablet. For naming the event, the participants typed in their response. We recorded every movement of pictures, their final sequence responses, and their naming responses. The task was self-paced, and participants could move around the pictures as many times as they wished, until they were satisfied with the sequence. The task took between 15 and 35 min, depending on individual participants’ pace.

Statistical analyses

The recorded macro-event names were scored by two independent assessors. One of them was blinded to the diagnosis of the participants. Any discrepancies between assessors were discussed until a conclusion was reached. Sequence accuracy was measured using edit distance (for details, see Results).

All analyses were conducted in R, version 4.0.4 (R Core Team, 2021). The Shapiro–Wilk test, along with visual examination of density plots, was used to determine normality of distribution, and Levene’s test was conducted to check for heteroscedasticity. For outcome variables that were not normally distributed, the nonparametric Kruskal–Wallis test was conducted, followed by Dunn’s multiple comparison test, and a Benjamin-Hochberg correction for multiple comparisons. For normally distributed variables with unequal variances, a Welch’s ANOVA was conducted (Delacre, Leys, Mora, & Lakens, 2019), followed by Games–Howell post hoc test. A robust factorial ANOVA using 10% trimmed means was conducted using the ‘WRS2’ package (Mair & Wilcox, 2020) to analyse the effect of cueing and diagnosis on sequence accuracy. Correlations were examined using Spearman’s rho correlation coefficient. The Levenshtein distance was calculated using the ‘stringdist’ package (Van der Loo, 2014).

Results Naming accuracy

Naming accuracy was measured by scoring participants' responses, as correct or incorrect, that is, ‘1’ or ‘0’, for each trial. Acceptable responses were the predetermined event terms, synonyms, or alternate terms that encapsulated the macro-event (e.g., for the event ‘eating at a restaurant’, ‘restaurant visit’, or ‘going to a restaurant’ would be acceptable responses). Naming or descriptions of individual micro-events, or specific objects within the pictures were scored as incorrect (e.g., for the event ‘grocery shopping’, ‘paying’, or ‘making a list’ would be unacceptable as they are not indicative of the event as a whole, but single pictures/actions within the event). These general categories of acceptable or unacceptable responses were predetermined, but not specific responses. Percentage accuracy was calculated for each participant. The means are provided in Table 1.

Table 1. Demographic characteristics and performance on measures in Experiment 1 (Means and SD) Measure OA (n = 20) MCI (n = 10) AD (n = 10) p a η 2 Age 70.9 (5.5) 72.6 (6.2) 74.2 (7.5) .17 .04 Sex (F/M) 16/4 3/7 7/3 .02b .43c Education (years) 11.8 (1.7) 10.4 (1.8) 9.9 (1.7) .01 .18 MMSE score (max. 30) 28.9 (0.9) 24.7 (1.3) 23 (1.9) <.001 .76 Naming accuracy (%) 93.6 (7.3) 61.4 (27.8) 55.7 (31.2) <.001 .51 Sequence accuracy (LDd) 0.25 (0.2) 0.69 (0.59) 1.03 (0.55) <.001 .35 Average no. of moves per trial 4.13 (0.1) 4.3 (0.2) 4.56 (0.4) <.001 .34

Commonly occurring errors in naming included an inability to come up with a response, describing each micro-event instead of naming the macro-event, naming just an object within the pictures, or naming unrelated event terms. One commonly observed error in the patients was their preoccupation with one particular presented event and use of that event term repeatedly for subsequent event trials, even when the events were unrelated.

To analyse the differences between the groups formally, the nonparametric Kruskal–Wallis was conducted. The test revealed a significant effect of diagnostic category on naming accuracy, H(2) = 21, p < .001, η2 = .51. Further examination showed that AD and MCI groups performed significantly worse than the OA group, but did not differ significantly from each other (see Figure 2).

Group-wise means on Experiment 1 measures of (a) naming accuracy; (b) sequence accuracy; (c) average no. of moves per trial (error bars represent SE).

Sequence accuracy

Sequence accuracy was calculated using the edit distance (‘Levenshtein distance’). In this, observed and expected picture sequences were represented as four-letter strings. Expected sequences always had the format 'ABCD' (normed order of the four sub-events). Observed sequences were coded using the same four letters. The order of letters corresponded to the order of the sub-events, as arranged by the participants. Observed and expected sequences from each participant on every trial were then compared by calculating how many edits of the observed sequence were necessary to derive the expected sequence. For example, the response ‘CADB’ would require four edits, and, therefore, is more different than the response ‘BACD’, which requires only two edits. A higher number of edits indicated a greater distance from the original string. The lowest distance value was '0' (observed sequence was identical to the expected sequence). The highest distance value was '4' (all pictures in the wrong position). The distance values were averaged across trials (for means, see Table 1).

A Kruskal–Wallis test, conducted to analyse group differences, revealed a significant effect of diagnostic category on sequence accuracy, H(2) = 14.98, p < .001, η2 = .35. The AD and MCI groups had lower sequence accuracy (i.e., higher Levenshtein distance) compared with the OA group, but did not differ significantly from each other. Additionally, we observed a strong correlation between naming and sequence accuracy, ρ = −.61, p < .001, such that naming accuracy was lower when Levenshtein distance was greater.

Moves

We recorded how many moves participants made on each trial before finalizing their response, and then calculated the average per trial for each participant (see Table 1). A Kruskal–Wallis test showed that the number of moves was influenced by diagnostic category, H(2) = 14.7, p < .001, η2 = .34. The OA group made fewer moves compared with the AD group. The MCI group did not differ from either AD or OA groups. Further, the number of moves correlated negatively with naming accuracy, ρ = −.49, p = .001, but the correlation with sequence accuracy was unclear, ρ = −.31, p = .05.

Effect of cueing

Sequence accuracy was examined separately on cued and non-cued trials. The 10% trimmed means for each group on cued and non-cued trials are provided in Table 2. A robust 3 (diagnosis: OA vs. MCI vs. AD) × 2 (trial type: cued vs. non-cued) mixed-factor ANOVA revealed a main effect of diagnosis, Ft(2, 14.3) = 8.93, p = .005. The main effect of trial type and the interaction between diagnosis and trial type were not significant Fts < 1, ps > .85. Further, sequence accuracy, specifically on non-cued trials, was strongly correlated with naming accuracy, ρ = −.70, p < .001, with higher naming accuracy associated with a smaller Levenshtein distance.

Table 2. Group-wise performance on cued and non-cued trials in Experiment 1 (10% trimmed means and 95% confidence intervals) OA MCI AD M CI M CI M CI Sequence accuracy (Levenshtein distance) Cued 0.2 [0.11, 0.3] 0.64 [0.29, 1.05] 0.98 [0.61, 1.38] Non-cued 0.25 [0.14, 0.41] 0.59 [0.25, 1.16] 1.0 [0.57, 1.5] Average moves/trial Cued 4.1 [4.0, 4.2] 4.3 [4.1, 4.5] 4.3 [4.1, 4.8] Non-cued 4.1 [4.0, 4.3] 4.2 [4.1, 4.4] 4.7 [4.4, 5.0]

Effect of cueing was also examined for the average number of moves per trial. The group-wise 10% trimmed means on cued and non-cued trials are provided in Table 2. A robust 3 (diagnosis: OA vs. MCI vs. AD) × 2 (trial type: cued vs. non-cued) mixed-factor ANOVA was conducted. Similar to sequence accuracy, we only found a main effect of diagnosis, Ft(2, 11.3) = 10.8, p = .002; all other effects were not significant, Fts ≤ 3.21, ps ≥ .07.

Discussion

This experiment demonstrated that AD and MCI populations have difficulty in establishing a temporal and causal order of visually depicted four-event sequences. These findings reflect deficits in the activation and the use of an appropriate macro-event schema which serves integration of sub-events. This was evident in the low sequence accuracy rates displayed by the two groups, as well as in naming the depicted macro-events appropriately. These findings are in line with previous studies that demonstrated deficits experienced in binding related pieces of information in AD patients but not in healthy older adults (Parra et al., 2009; Parra, Abrahams, Logie, & Della Sala, 2010) or non-AD dementias (Della Sala, Parra, Fabi, Luzzi, & Abrahams, 2012). Additionally, MCI and AD groups also displayed more uncertainty in finalizing the sequence, as evidenced by the higher number of moves on average per trial.

The low naming accuracy among the patient groups may, to some extent, be attributed to anomia experienced during AD. However, as is suggested by a lower sequence accuracy in addition to lower naming accuracy, the difficulties experienced by AD and MCI patients appear to emerge in part by deficits in non-verbal aspects of event cognition. Our findings indicate that individual pictures are not readily recognized as depicting interrelated sub-events and, therefore, do not serve as cues to activate an appropriate macro-event schema, on the basis of which the appropriate temporal sequence could be established and an appropriate name could be retrieved. As the individual sub-events can be considered as segments of what is a continuous flow of information, one explanation for the results may be that subjects are not able to infer what is not depicted and, therefore, cannot find the links between sub-events.

Cueing with the event name did not improve sequence accuracy. In the cognitively healthy older adults, this may be because their performance was at ceiling. There may be several reasons for the lack of an effect in the two patient populations. One, patients might not be able to access sub-events of the macro-event, that is, they may not be able to access a more fine-grained level of event representation. Two, the issue might be an inability to associate the individual pictures with the schema that was activated by the verbal cue.

Working memory deficits which are commonly attested for in AD and MCI individuals cannot explain the effects of ordering, because the items to be arranged, as well as the cue, were visible throughout the trial and did not need to be held in memory. In sum, the performance of the AD and MCI group points to impaired bottom-up processing – retrieving the macro-event schema from the sub-events – as well as top-down processing – activating sub-events from the macro-event schema (verbal cue).

Experiment 2

In Experiment 1, both patient groups were evidently impaired in picture sequencing and subsequent macro-event recognition. Experiment 2 aimed to extend these findings to investigate whether macro-event recognition is impaired in AD and MCI when the additional cognitive load of unscrambling event sequences was reduced. Similar to Experiment 1, participants were presented four pictures depicting sub-events. The pictures were presented in a temporally and causally appropriate sequence, but in a staggered form. Participants were to stop the trial when they thought they could identify the macro-event and then name it. The goal was the activation of the macro-event being depicted in as few pictures as possible, partly by drawing causal inferences between the micro-events that are available and partly by predicting the micro-events that would follow.

Method Participants

The same individuals who participated in Experiment 1 also participated in Experiment 2. In the cognitively healthy group, in addition to the twenty participants from Experiment 1, data from two additional participants, who were excluded from Experiment 1 due to a technical error, were also included. A total of twenty-two cognitively healthy older adults were included in this experiment. Demographic characteristics are reported in Table 3. The MCI and AD groups remained the same as in Experiment 1.

Stimuli and procedure

The stimuli consisted of fourteen sets of events that were not used in Experiment 1. Each set consisted of four pictures, each showing different sub-events of one macro-event. Each picture was 300 pixels in width and height. The experiment was programmed using JavaScript and was presented on a 10.1″ tablet with an external keyboard. The pictures were presented in the correct sequence, and each trial began with the presentation of the first two pictures in the sequence. After a 5-s interval, the third picture was presented. This was followed by another 5-s interval before the final picture appeared. Participants were instructed to press ‘space’ or touch the screen as soon as they were able to identify the macro-event being depicted, to end the trial, and then write down the name of the event. It was emphasized that participants were to name the macro-event, and not the individual micro-events. The trials were presented in a randomized order. Participants were given five practice trials.

We measured the number of pictures seen by the participants before identifying the macro-event, as well as the naming accuracy. Naming accuracy was scored by two assessors independently, one of whom was blinded to the diagnosis of the participants. Any discrepancies in scoring were resolved with discussion.

Results Naming accuracy

The naming responses were scored as ‘1’ or ‘0’, for a correct and incorrect response, respectively, and the percentage of correct responses was calculated (for means, see Table 3). A Kruskal–Wallis test revealed a significant effect of diagnostic category on naming accuracy, H(2) = 21, p < .001, η2 = .49. Further multiple comparisons indicated significantly lower naming accuracy in the AD and MCI groups compared with the OA group, but no difference was observed between AD and MCI groups.

Table 3. Demographic characteristics and performance on measures in Experiment 2 (Means and SD) Measure OA (n = 22) MCI (n = 10) AD (n = 10) p a η 2 Age 71 (5.3) 72.6 (6.2) 74.2 (7.5) .16 .04 Sex (F/M) 17/5 3/7 7/3 .03b .40c Education (years) 11.9 (1.6) 10.4 (1.8) 9.9 (1.7) .006 .21 MMSE score (max. 30) 28.9 (0.9) 24.7 (1.3) 23 (1.9) <.001 .75 Naming accuracy (%) 97.0 (5.3) 70.7 (28.1) 81.4 (8.4) <.001 .49 Avg. no of pictures viewed per trial 2.24 (0.2) 2.79 (0.5) 2.83 (0.5) .003d .28 Pictures

We calculated the average number of pictures that each participant viewed per trial before registering their response. The group-wise means are presented in Table 3. A Welch’s ANOVA revealed a significant effect of diagnostic category on the number of pictures viewed, FW(2, 13.2) = 9.65, p = .003, η2 = .28. Games–Howell post hoc test indicated that AD and MCI groups viewed more pictures on an average compared to the OA group. AD and MCI groups did not differ significantly from each other (see Figure 3). Additionally, we observed a significant correlation between naming accuracy and number of pictures viewed, ρ = −.53, p = .003, indicating higher accuracy when fewer pictures were viewed.

Group-wise means on Experiment 2 measures of (a) naming accuracy; (b) average no. of pictures viewed per trial (error bars represent SE).

Discussion

This experiment sought to extend findings from Experiment 1 to investigate macro-event recognition, when, in contrast to Experiment 1, recognition was facilitated by sequential sub-events, thereby reducing cognitive load. Naming accuracy was significantly lower in MCI and AD groups compared to the cognitively healthy adults, but the MCI and AD groups did not differ from each other. Naming accuracy was lower in patient groups despite the fact that they, on average, viewed significantly more pictures per trial than the cognitively healthy group, pointing to an inability to integrate information even when it is presented without distortion. In Experiment 1, there was an added challenge of unscrambling the pictures before participants named the macro-event. Here, even when this hurdle is not present, the patient groups still display difficulty in macro-event identification. That correct event sequences did not facilitate recognition indicates that macro-event recognition may not necessarily be dependent on correct picture sequencing, rather correct sequencing relies on event recognition.

The task also involves some predictive inferencing. In order to identify the macro-event being depicted, while not viewing all sub-events, would require predictive processing. Prediction is implicit to visual perception (Cohn, 2019; Enns & Lleras, 2008) and is based on existing schemata for events (Zacks et al., 2007). It is essential for higher efficiency in information processing, but is impaired in neurodegenerative diseases, such a