The Structure of Memory and its Role in Envisioning the Future: A Network Study
Emma E. Schreurs University of Amsterdam
Abstract
Memory theorists posit that future imagination entails the construction of bits and pieces of information from episodic memory into an imagined future event. The goal of this pilot study was to assess the features and the structure of the features that compose a memory or future imagination. First, it was aimed to test whether people draw on the same features to imagine the future as to remember the past. Second, it was aimed to test whether people draw on the most central features of the episodic memory network to imagine the future. Twelve
participants filled in an online survey in which they described 100 memories and 20 future imaginations. Features were the words used in the descriptions, the locations and the people included in the event. An episodic memory feature network was created based on a bipartite network that consisted of features and their connections to memories. The aims were tested using the episodic memory feature network. Results showed that on average 47% of the features used to imagine the future were also used to remember the past. Also, it showed significant results on the t-test comparing the centrality of episodic memory features that were also in an imagined future event with the centrality of the features that were exclusively in the episodic memory network. Together, the findings from these specific aims provide support for the notion that when we imagine the future we draw on the same pool of features that we draw on to remember the past. Moreover, it suggests that there is a structure to the pool of features used during remembering the past which seems to affect what we will imagine will occur in the future, for we tend to draw on the most central features of the episodic memory network to imagine the future.
Introduction
Memories enable human beings to remember past experiences. Memory can be divided into numerous categories, of which two commonly used are semantic memory and episodic memory. Semantic memory consists of general knowledge about the world attained through the media, education, and other sources of information. It includes memory for faces, melodies, topography, knowledge of language, and so on (Tulving, 1985; Yasuda, Watanabe, & Ono, 1997). Episodic memory, the focus of this study, is defined as memory for personal experiences, which are dated and located in subjective time and space (Tulving, Schacter, McLachlan et al., 1988). It allows individuals to project themselves backward in time and recollect many aspects of their previous experiences (Tulving, 1985).
Episodic memory is thought to be a reconstruction of pieces of information from one’s past and not a direct reproduction. This concept is called the constructive brain. For example, remembering that you fell from your bike is a reconstruction of bits and pieces of information which are stored in the brain and not a literal reproduction of the event. There is considerable evidence to support this concept. For example, false recognition errors are common in word recall procedures when the target word (e.g., sleep) is closely connected to words that were presented (e.g., bed and pillow) relative to when there is no such connection with the words presented (e.g., chair and rain; Schacter & Addis, 2007). Fewer false-recognition errors are made when someone has brain damage to the hippocampus, a part of the brain that is required for episodic memory recall (Cabeza et al., 2001). The fact that more false-recognition occurs when episodic memory is intact, suggests that functioning episodic memory involves
reproduction based on concepts (e.g., sleep) rather than literal reproductions. Similarly, in the domain of episodic memory, false memories are common in some patients. The mere fact that we can have a memory that we are convinced occurred but did not, strongly suggests that memory is not a reproduction but a reconstruction. False memories cannot occur in a system
that just reproduces (Schacter & Addis, 2007; Burgess & Shallice, 1996; Schnider, 2003). Together, these studies give evidence for constructive processes rather than reproductive processes.
Memories comprise different features. Commonly used features are: what, where and when (Tulving, 1983). The “what” can stands for a couple of things. First of all, it can stand for the question: “what happened?” so that a person can describe the event of the memory. Next to this, it can include the people involved in the memory and the emotions felt. The “where” stands for the question: “where it happened?”, so that a person can describe the location of the memory. The “when” stands for the question: “when it happened?”, so that a person can describe the date or period in which the memory took place. Together, the “what” and the “where” set the scene on which the recollected event took place, this is named ‘scene construction’ (Hassabis & Maguire, 2007). These features make up at least part of the larger pool of features upon which we draw to construct a memory.
Episodic memory has been studied for the past decades, focusing on recollection of the past (Tulving, 2002). But what if we start to think about the advantages of looking into the future? Consider history. One of the reasons we study this is to predict and imagine how the course of the future may turn. It sensibly seems impossible to predict or imagine the future without having any stored knowledge about life. People draw on past experiences in order to imagine and simulate events that might happen in their personal futures (Dudai & Carruthers, 2005). Looking into the future instead of into the past can serve as an exploratory concept into getting a deeper understanding of memory.
Future imagination is referred to as the ability to mentally project oneself in the future and thus the capability to pre-experience future events (Tulving, 1985, 2005). Like memory, future imagination is thought to be a constructive process, a hypothesis referred to as the “constructive episodic simulation hypotheses” (Schacter and Addis, 2007). This hypothesis
suggests that future imagination requires a system that recombines details from episodic events. It is even suggested that one of the main functions of memory is to make information available for future imagination. As previously noted, patients with damage to the
hippocampus have impairments of episodic memory. Less discussed is that these patients also have difficulties imagining future experiences (Rosenbaum et al., 2005). Next to this, episodic memory and future imagination are similar in that they both involve creating a mental image of personal events, which contain strong contextual components (Tulving, 1985). Last, it is said by Tulving (2001) that both episodic memory and future imagination are characterized by auto noetic consciousness, which enables an individual to be aware of one’s subjective experience during the mental simulation of an event. These similarities suggest that
individuals may draw on the same pool of features to remember the past as when to imagine the future. Up to now, no one has assessed these features and their relations to one another. Moreover, there is no existing definition of the term “features”.
Although there are definite similarities between memory and future imagination, there are differences too. In comparison to episodic memory, future imagination arguably involves more goal-directed processing. D’Argembeau, Stawarczyk, Majerus and Collete (2010) performed a study in which they investigated the activation of different brain areas during future imagination, with an emphasis on goal-directed processing. They found that a
particular brain area (MPFC) was very active during future imagination. This brain area has been associated with self-referential processing, emotional processing and decision making, but not with episodic memory retrieval. Therefore, goal-directed processing may be a critical difference between episodic memory and future imagination. Also, it has been suggested that future imagination in comparison to memory involves more cognitive control (Zheng, Luo, & Yu, 2014). Cognitive control is the ability to coordinate and structure thoughts and actions based on internal behavioral goals (Braver, 2012). Considering that this capability is based on
internal goals (Braver, 2012), it makes sense that cognitive control has a bigger function within future imagination than in memory. These findings suggest that individuals may not draw on the entire pool of episodic memory features to prospect but, instead, draw on those features most relevant to current goals.
Emotion of memory and future imagination also suggests differences in the process of remembering the past versus imagining the future. Painter and Kring (2014) performed a study in which they looked at the content and style of freely generated narratives of
participants positive, negative and neutral memories and future imaginations. They found that people are more likely to base positive future imaginations on memories and are less likely to do so for negative future imaginations. These findings suggest that when individuals draw on features from which to construct a future event, they draw especially from features derived from positive memories. Other studies have shown that positive memories and future imaginations are experienced more vividly than negative memories and future imaginations (D’Argembeau & Van der Linden, 2004; Rasmussen & Berntsen, 2013). Accordingly, these studies provide further evidence that we may draw more heavily on some specific features from episodic memory than others (i.e., those derived from positive memories and those most relevant to current goals).
The Current Study
In this research project, my colleagues and I aimed to develop a research method that can be used to investigate questions about episodic memory, future imagination, and the features that compose them. To do so, we used network modeling. Network modeling is data-based and represents objects and the relationships between them. Network modeling has been applied in various fields of research (e.g., Medicine, Biology, Psychology). Consequently, there is a wide range of different network models. However, this method had not yet been applied to the representations and description of memories, future imaginations and their
features. This method may also give insight into the nature of this pool of features we use to imagine the future and how the structure of relationships among those features might be informative as to how someone imagines the future.
For this thesis, I was especially interested in the features that individuals use for memory and future imagination. I used the information of the networks to examine two central ideas. First, I aimed to investigate whether people indeed draw on the features of the episodic memory network to imagine future events by examining the portion of features in the future imaginations that were also in the episodic memory network. Second, I investigated whether people draw especially on features highly central in the memory feature network to imagine the future. The second idea is based on the following argument: Future imagination involves goal-directed processing (D’Argembeau, Stawarczyk, Majerus and Collete, 2010). In memory studies it had been suggested that this goal-directed nature is a defining characteristic of self-defining memories which are thought to have a high connectedness with other
memories (Conway, Singer & Tagini, 2004; see also Kehrer, 2018). Highly central memories likely contain highly central features in the feature network. Accordingly, it was hypothesized that people tend to draw on the most central features of the episodic memory feature network when they imagine the future.
Method Participants and Procedure
For this pilot study, 70 participants completed informed consent. 28 individuals
dropped out before completing a single question and 11 individuals started the first survey but did not finish. Leaving us with 31 participants. 17 participants dropped out before completing survey five, taking us to 14 participants. From these 14 participants, two were omitted
because they replied in Dutch, giving us our final sample of 12 participants. The participants were recruited through administration of the project to the lab of the University of
Amsterdam. On the website of the lab, there was a short description of our study, hereby people were able to read the project and decide whether to participate. The lab mainly aims for the recruitment of first year students who need to earn 20 participation points in the first year of their studies. Therefore, 3.75 participation points were awarded for participation. The participants that applied to the survey could immediately click on a link to start the survey. At the beginning of the survey each participant read an information letter and was asked to sign an informed consent. It was made clear to each participant that participation would be completely voluntary and anonymous.
Each participant was sent an email in the following days of participation with a link to the survey. Also, a reminder email would be send every morning, so the participant would not forget to fill in the survey. The survey was to be taken into five different sessions of which each session took approximately an hour. The participants were encouraged to finish the survey within five days, one session each day. In every session the participant was asked to describe 17 episodic memories, three self-defining memories, four future imaginations, and were to complete several questions about those events. If a participant had any question about the survey it was possible to email, and a response would be given as soon as possible.
Materials
The survey was made with a programme named Qualtrics (Qualtrics, 2013). We designed a survey which consists of different blocks with each their own information sheet and/or list of questions. First, a description was given of episodic memory and the task which was going to be required in this category. 17 episodic memories were to be described, with six sub-questions for each memory. For example, one of these sub-questions was: “What
city/country did this event take place in? If you don't know any piece of information, leave blank”. The information conveyed in these sub-questions were the approximate date of the event, location of the event, people in the event and emotions felt. The second task of the survey started off with a short description of self-defining memories. Three self-defining memories were to be described, with each 11 sub-questions. Six of the sub-question were the same as for the episodic memories. In addition, five sub-questions were added specific to self-defining memories. Before the final task of the survey, a short description was given about future imagination. Four future imaginations were to be filled in, with each seven sub-questions. Six sub-question were adapted from the episodic memory questions in order the describe a future imagination. One sub-question was added; “To what extent does this imagined future event tell us something about who you are as a person?”.
Analysis Plan
Data analysis was performed in the statistical platform R (R Core Team, 2013). For each of the participants two personal networks were made: an episodic memory network and an episodic memory feature network. To make these networks, we completed three steps on each individual memory and future imagination (see Table 1). The first step was to separate every description of a memory or future imagination into individual components and to modify the words to lower case and free of punctuation. This was done with R-package “tokenizers” (see Table 1, step 1). This was an essential step, for we wanted to use the
important words of every single description as features of memories or future imaginations. The second step was to remove unwanted components. This was done to filter out the words that did not provide important information to the memory or future imagination. The
following components were removed; stop words, amplification words, adverbs, adjectives, auxiliary verbs, complex prepositions, prepositions, pronouns and function words. We used libraries of words in R to remove most of the unwanted components, R-packages
(“stopwords”, “tm”, “qdap”, “qdapDictionaries”). However, not all components could be removed with existing libraries in R, therefore we made lists of the auxiliary verbs, adjectives, complex prepositions and prepositions (“What is an Adjective,”2018; Essberger, 2012; “Lists of pronouns in English,” 2018; “Auxiliary verbs,”2018). Eventually, the libraries and lists of unwanted components were removed in R with package (“tm”, function “removeWords”) (see Table 1, step 2). The third step was to combine the remaining components into a vector of words, each vector belonging to a different memory. This was then made into a matrix consisting of two columns. One column that specified the memory or future event (e.g., M1) and one column where all of the components belonging to the memory were listed (see Table 1, step 3). Also, the specific location and the persons partaking in each memory and future imagination were added to the list of components. This process was repeated for all memories and future imaginations, providing a table that specified every connection between a feature and a memory or future imagination (edge-list).
Table 1
The data preparation steps
Description of a memory
I was in the Amsterdam Forest with my parents on my bike with training wheels. My parents decided to take of the training wheels for once, I immediately jumped on my bike
and rode away. I was very excited.
“i” “was” “in” “the” “amsterdam
”
“forest” “with” “my”
“parents ”
“on” “my” “bike” “with” “training
” “wheels” “my” “parents ” “decided ’
“to” “take” “of” “the” “training
”
“wheels ” “for” “once” “i” ‘immediately
”
“jumped” “on” “my” “bike”
“and” “rode” “away ”
“i” “was” “very” “excited
”
“amsterdam” “forest” “parents” “bike” “training” “wheels” “parents” “decided” “take” “training” “wheels” “immediately”
“jumped” “bike” “rode” “away” “excited”
M1 “amsterdam” M1 “forest” M1 “parents” M1 “bike” M1 “training” M1 “wheels” …… ……. Step 1 Step 2 Step 3
The matrix that was created in the final step was used as an edge list. An edge list encodes every relationship present in a network. For this study, it encoded every single relationship that a memory has with a feature. We used these edge lists to calculate our bipartite networks. A bipartite network is a network that has two sets of nodes (U and V). In our case, the set of U nodes are the memories and the set of V nodes are the features. Features are connected to a memory if the feature was included in that memory. As depicted in Figure 1, memories (M1, M2, etc., the U-nodes) never connect with each other, only with features (F1, F2, etc., the V-nodes). Thus, nodes in the U-set can never connect to another node from
U-set, only to V nodes.
Figure 1. Example Bipartite Network. M=Memory. F=Feature. Red boxes signify memories. Blue circles signify features
The bipartite network can be used to create two network projections. An example based on Figure 1 is visualized in Figure 2. These projections were made with the R-package “igraph”. Igraph is able to create the bipartite network based on ones’ individuals edge list. The
bipartite network is depicted in the middle of Figure 2. Projection U represents a feature network based on its shared memory connections (Figure 2). Projection U demonstrates that F3 and F5 are not connected, this is because they do not have a shared connection to a memory. Projection V represents a memory network based on its shared feature connections
(Figure 2). Projection V demonstrates that all the memories are connected, because they all have shared features (as seen in the middle of Figure 2).
Figure 2. Example bipartite network; Two visual projections of networks based on shared U-nodes=Projection V. And shared V-U-nodes=Projection U.
In this project, I will focus on the episodic memory feature network. I will describe this network reporting its mean number of nodes, mean number of edges, mean density and the mean of the average degrees. Nodes are the components of the network and edges are the direct interactions between the nodes. The mean of the nodes is calculated as the sum of the nodes divided by the number of participants. The same holds for the edges. Hereafter I calculated the mean density of the twelve networks. The density describes the portion of potential connections compared to the actual connections of the network. A “potential connection” is a connection that could exist between two nodes, regardless of whether it actually does. The density was calculated in R with package (“igraph, function is
“edge_density”). If a network has a high density, then almost all of the nodes are connected. Final, I calculated the mean of the average degree of the nodes, which is the number of edges
connected to the node. Degree was calculated in R with package (“igraph, function is “degree”). Average degree is the average across all nodes in the network.
To answer the first research question; whether people draw on the features of the episodic memory network to imagine future events, I calculated the proportion of features in the imagined future events that were also in the episodic memory feature network. Secondly, to test the hypothesis that people tend to draw on the most central features of the episodic memory feature network to imagine the future, I compared the centrality of episodic memory features that were also in an imagined future event with the centrality of the features that were exclusively in the episodic memory network. To do so I used a simple t-test on four properties of centrality; degree, strength, betweenness and closeness. The degree was previously
described. The strength of a node is the total weight of the edges belonging to the node. (Barrat, Barthelemy, & Vespignani, 2007). Betweenness is the number of times a node acts as a bridge along the shortest path between two other nodes. The closeness is the inverse of the average length of the shortest path between a node and all other nodes in the graph. The more central the node, the closer it is to all other nodes (Newman, 2001). All centrality properties were calculated using R package (“igraph”). Final, the most central features of the networks were deidentified (to maintain anonymity) and evaluated on content (e.g., location, people, emotion).
Results
First, the mean and the standard deviation of the number of nodes and edges for the twelve participants are calculated from the episodic memory feature network. The mean of the nodes is 716.33 with a standard deviation of 162.85. The mean of the edges is 9147.92 with a standard deviation of 3480.29. The number of nodes and edges varied widely across
individuals, with a range of 401-990 for nodes and 2540-14576 for edges. Figure 3 visualizes the feature network for two participants at the extremes of these properties. For participant 10, there are many nodes (990) and many edges (14576) among those nodes. In contrast, for participant 1, there are fewer nodes (401) and fewer edges (2540) among those nodes.
Second, the mean and the standard deviation of the density and the average degree are calculated. The mean of the density is .03471 with a standard deviation of .00372. The
standard deviation is small suggesting the densities of the networks are approximately equal to each other. The mean of the average degree is 24.63 with a standard deviation of 5.10. The average degree varied across individuals, with a range of 13.17-29.45. Figure 3 visualizes that for participant 10 the average degree (29.45) is high. In contrast, for participant 1, the average degree (13.17) is low. In conclusion, the episodic memory feature network can vary widely across individuals in number of nodes, edges and the average degree.
Figure 3. Episodic memory feature networks of two participants. Left visualization is of participant 10 and right visualization of participant 1. The node (features) sizes and colors (light to dark red) are based on its degree. The edges are colored blue.
For the first research question, the proportion of features in the imagined future events that are also in the episodic memory feature network are calculated for each participant (see Table 2.). These proportions are consistent with a mean of (𝑥𝑥̅ = .47) and standard deviation of (σ = .036)
Table 2.
Proportion of feature overlap between the imagined future events and the episodic memory feature network. Calculated for each participant.
To test the second hypothesis, a two samples t-test was performed on the following centrality properties; degree, strength, betweenness and closeness. Herein the features that were both included in the imagined future events and the episodic memory network were compared with the features that were exclusively in the episodic memory network, thus not in the imagined future events. Table 5 displays the means of every centrality property for the features that are exclusively in the episodic memory network, the means of the features that are in both the episodic memory network and the imagined future events, and the means of the Effect-size (Cohen’s-d). It is clear that means of each property differ largely between the two groups (see Table 5). Subsequently, all t-values are lower than -2.00, all the p-values are lower than .001 and all effect sizes are higher than .80. There is only one exception
(participant one betweenness: t = -0.86, p = .39. In conclusion, there is a significant difference and a large effect size for the degree, strength, betweenness and closeness of the
Participant Proportion 1 .42 2 .45 3 .49 4 .45 5 .46 6 .47 7 .40 8 .47 9 .49 10 .49 11 .51 12 .51
features included in the imagined future events compared to the unincluded features. Figure 4 visualizes the centrality of the features that are were both included in the imagined future events and the episodic memory network (i.e., those in blue) and of the features that were exclusively in the episodic memory network (i.e., those in red), for one participant. Here it is seen that the nodes used in future events (i.e., those in blue) tend to be among those with the highest centrality (i.e., the largest nodes). In conclusion, the features that are central to the episodic memory network occur in the imagined future events.
Figure 4. A visualization for one individual of the centrality of features that are exclusively in the episodic memory feature network and of features that also appear in the imagined future events. Nodes are the features in the episodic memory network. The larger the node, the higher the centrality. If nodes are “stacked” it means they have the same centrality. Features
Features that also appear in the imagined future events.
Features that are exclusively in the episodic memory network
Final, all central features were evaluated and deidentified. Table 3 displays the most central features of participant 10, that are in both the episodic memories as future
imaginations. The words demonstrate the content of the most central features. Most of the features are locations and people. Table 4 displays the most central features of participant 10, that are exclusive to the episodic memory network. Here you can see that most of the features are verbs; less people and location are used.
Table 3.
Features most central to the episodic memory network, that were also used to imagine the
future. Participant 10.
Table 4.
“person1” “walking” “look” “my house” “people”
“went” “took” “talking” “staying” “person7”
“house” “brother” “time” “morning” “middle school”
“school” “mom” “person4” “ran” “university”
“sitting” “water” “running” “playing” “person4”
“day” “end” “grade” “reading” “stood”
“person2” “going” “middle” “spent” “said”
“dad” “at home” “christmas” “laughing” “ended”
“go” “night” “person5” “making” “wanted”
“friend” “friends” “walked” “person6” “way”
“elementary”
“school” “told” “person3” “lunch”
“class” “tree” “found” “dinner”
“working” “hair” “teacher” “come”
“front” “gym” “set” “book”
“girl” “new” “store” “person1”
“family” “hiking” “fell” “café”
“looking” “kept” “later” “game”
“parents” “kids” “picture” “metal”
Features most central exclusively to the episodic memory network. Participant 10.
“trip” “sleeping” “asked” “take”
Table 5.
Means for the four centrality properties for each participant (P) of the features that are in the episodic memory feature network (EM) but not used in an imagined future event and the features that are both in the episodic memory feature network and were used when imagining a future
Degree Strength Closeness Betweenness
P EM EM and FI d EM EM and FI d EM EM and FI d EM AM and FI d
1 10.86 (9.53) 24.09 (25.11) .98 11.52 (12.00) 28.90 (33.85) .98 0.00025 (3.45e-05) 0.00026 (3.25e-05) .11 173.60 (616.61) 1109.57 (2014.08) .93 2 17.60 (10.56) 45.31 (44.06) 1.35 18.15 (11.88) 51.40 (54.98) 1.32 0.00032 (2.31e-05) 0.00033 (1.92e-05) .75 170.23 (459.05) 2248.70 (4585.97) 1.07 3 18.67 (13.08) 43.78 (47.22) 1.00 19.96 (17.23) 54.15 (68.36) .95 0.00061 (4.61e-05) 0.00066 (6.51e-05) .92 140.77 (485.58) 1606.93 (3878.63) .77 4 16.11 (11.27) 46.14 (44.83) 1.44 16.95 (14.59) 57.14 (63.90) 1.39 0.00071 (5.37e-05) 0.00077 (8.93e-05) .98 143.14 (475.29) 1784.87 (3314.90) 1.17 5 16.66 (12.11) 39.07 (41.64) 1.08 17.61 (14.63) 48.01 (62.69) 1.02 0.00074 (6.80e-05) 0.00080 (9.53e-05) .79 183.24 (649.48) 1403.88 (3054.85) .86 6 26.44 (26.00) 53.48 (47.18) 0.88 27.89 (34.29) 59.54 (59.25) 0.80 0.00046 (3.72e-05) 0.00049 (4.66e-05) .95 375.18 (2119.51) 1831.46 (3574.10) .60 7 20.93 (13.26) 46.77 (46.85) 1.16 21.51 (14.64) 52.28 (59.63) 1.13 0.00047 (3.35e-05) 0.00050 (5.90e-05) .96 274.96 (876.23) 2618.87 (4996.01) 1.10 8 22.68 (14.48) 64.29 (63.32) 1.38 24.02 (17.92) 78.92 (94.62) 1.25 0.00058 (4.26e-05) 0.00062 (6.15e-05) .95 136.41 (348.03) 1987.92 (5250.16) .81 9 18.90 (13.44) 47.92 (56.17) 1.21 19.56 (15.17) 56.57 (75.36) 1.20 0.00055 (4.33e-05) 0.00059 (6.28e-05) .88 214.49 (730.54) 2709.67 (6507.54) 1.02 10 24.03 (16.82) 58.64 (56.98) 1.27 24.69 (18.89) 65.43 (71.39) 1.23 0.00042 (3.02e-05) 0.00045 (4.74e-05) 1.10 301.32 (1096.09) 2853.56 (5967.87) 1.00 11 22.10 (19.07) 51.85 (61.21) .96 23.49 (24.77) 62.75 (90.79) .89 0.00062 (4.67e-05) 0.00065 (6.42e-05) .64 187.52 (709.48) 1696.05 (5210.57) .67 12 22.87 (16.38) 52.66 (59.86) 1.04 24.01 (19.49) 62.80 (83.61) 1.01 0.00058 (5.03e-05) 0.00062 (6.82e-05) .74 202.82 (592.72) 1938.77 (4444.70) .91
Discussion
The first aim of this pilot study was to test whether people indeed draw on the features of the episodic memory network to imagine future events. This was done by examining the portion of features in the imagined future events that were also in the episodic memory network. Of the features used to imagine the future consistently around half were also used to remember the past. The number of overlapping features could be even higher if more memories were asked. In this study, 100 episodic memory descriptions were asked of each individual, which is just a small percentage of the amount of memories a person has in total. If we increased the number of memories, we would have a larger number of memory features and, thus, it would be more likely that the future imagination features will be found in the episodic memory feature network. Hence, I believe that these results constitute some evidence that people are drawing on the same pool of features to imagine the future that they do when remembering the past. Nevertheless, about half of the features used to imagine the future were not used to remember the past. One possible explanation, as previously discussed, is the fact that we did not ask for enough memories. Another possible explanation is that people may also utilize information that does not directly originates from a personal experience to imagine a future event (e.g., I have never been married, but could imagine myself having a husband one day).
The second aim was to test the hypothesis that people tend to draw on the most central features of the episodic memory feature network when they imagine the future. Features used during future imaginations were consistently significantly more central in the memory feature network than were features not used during future imagination. In other words, people indeed seem to draw on the most central features of the episodic memory feature network when they imagine the future. This could indicate that when people imagine the future, they do so base on features that often occur in ones’ memories and thus are connected to many other features. Together, the findings from these specific aims provide support for the notion that when we
imagine the future we draw on the same pool of features that we draw on to remember the past. Moreover, it suggests that there is a structure to the pool of features used during remembering the past which seems to affect what we will imagine will occur in the future. For example, if a memory contains a couple of features that are very central in its episodic memory feature network, it is very likely that this memory, and thus its features, have some influence as to what you imagine yourself doing in the future.
The findings from these two specific aims provide strong support for the “constructive episodic simulation hypothesis” (Schacter and Addis, 2007). Their hypothesis suggested that future imagination requires a system that recombines details from episodic events. Therefore, features should be the same for episodic memory and future imagination, and I found that around half of the features in a future imagination indeed already appeared in past events. Moreover, I was able to provide evidence that we use a structure to draw on the most central features, which are details (e.g., an emotion, a person, the location) from an episodic event, to imagine the future. This provides direct evidence for the “constructive episodic simulation hypothesis”. In addition, these findings provide information about the content of the systems structure next to just suggesting that such a reconstructing system exists. People do not draw on an open pool of features, there is a structure in the feature network that tells us something about how they imagine the future. The results showed that the features that occur in future imaginations tend to be the most central features of the episodic memory network. This implies that, in theory, we could take an individual, ask them to provide 100 episodic memories and then predict what they think will occur in their future. Future research can be done with our method, to investigate whether the structures within the feature networks hold for larger samples of memories and, thus, provide more knowledge about the construction of memories and future imagination.
In this pilot study, we generally aimed to create an episodic memory network based on its features. This had never been done before and we were able to create informative networks for each individual. One of the advantages of creating networks was that it advanced an explicit definition of connectedness between features based on its memories and vice versa. The possibility that this connectedness did not exist, gave it the significant scientific
advantage of being refutable. Another advantage was, that it gave us the opportunity to test hypothesis about possible structures within the networks and therefore get a deeper
understanding as to how memory is constructed, as supported by the results of the two previously discussed specific aims. However, there are limitations. First, between individuals there was a lot of variation within the number of nodes, edges and the average degree. It is unclear, whether this reflects genuine individual differences or differences resulting from how the features were assessed. If this variation comes from individual differences, it can be explained by the fact that people are different. For example, one person needs more words and finds more words of importance to explain their memories than another person might need. Features were among other things, words drawn from the description of the memories. Therefore, some people may use more features, thus have more nodes. Another explanation is, that some people may have hurried through the survey and therefore used less words in their descriptions. In theory, one could specify the number of features to be provided (e.g., give us only 10 features of the memory) but this would be imposing network properties that may not reflect genuine features of the network. Acknowledging the differences across individuals is of more importance than to impose limitations. If the variation results from how the features were assessed, it imposes a second limitation on this pilot study, i.e., what is a feature?
As discussed in the introduction, this study is the first to try and asses the pool of features and to advance a definition of when features and memories should be considered
connected. In doing so, it became clear that a definition of what constitutes a feature is of critical importance. For example, one could question what we imposed by saying what a feature is not (e.g., it’s not a stop word or an auxiliary verb) while not having any definition of what a feature is. Thus, the absence of a definition made it hard to make decisions about what is and what isn’t a feature. Memory studies have suggested that the “what”, “where” and “when” are features that people draw on to remember the past (Tulving, 1983). Based on these findings we decided to use the words describing the “what” and the “where”. Next to this, the “who” was added, for this was, in our opinion, of importance to a memory. We based the exclusion of some features on debating which features seemed not to contain valuable information about a memory or future imagination. Also, we specifically asked for the location of the event and the people in the event. The results showed, that the features that were most central to the networks and were in both the imagined future events as the episodic memory network, often contained information about the location and the people in the event. This may reflect, that these features are of high importance to the construction of a future imagination or memory. However, it could also be a result of specifically asking the
participants for the location and the people in the event. By doing so, we may have imposed network properties on the results. To test whether our decisions about which features to include and which not, questionnaires could be taken on a large sample of people, in which people could agree or disagree with the inclusion or exclusion of the features. As previously discussed, future researchers should come up with a definition of features, on which it could be based what features were to be included and what not.
Another limitation of this pilot study is the possibility of an artifact in the results. We based the features on the descriptions of the memories and future imagination and selected some unwanted components that we removed (e.g., stop words, auxiliary verbs, etc. see analysis plan). However, it is possible that we did not address all the components that needed
to be removed. As seen in the results, there are words as “go” or “end” that are very central in the network of participant 10. These words have an influence on the results, for they both occur as features in the imagined future events as in the episodic memory network. It is questionable whether these words are of importance to a memory or future imagination. Then again, if we had a definition of what a feature is, we could have included or excluded the words with more certainty. Also, some features could not be recognized as identical (e.g., end and ended, see results), for we did not have enough to time to figure out how to singularize and stem the words in R. Therefore, some of the feature nodes may have a higher degree than was reflected in our assessment. Future research, should try and figure out how to singularize and stem the words and include this in the R code. If a similar study would be performed with a much larger sample of people and memories, it is of importance to have completed the automation of the code with inclusion of singularizing and stemming of the words. As a person it is easy to identify ‘bike’ and ‘bikes’ as the same word. However, it would take more than a year to look through every feature of for example 1000 memory networks. A possible solution is to have coders create an automated script that could do this quickly.
This pilot study is the first to create and describe the networks of one’s episodic memories based on its features. The findings from my specific aims provided support for the notion that when we imagine the future we draw on the same pool of features that we draw on to remember the past. Next to this, it provided evidence that people indeed seem to draw on the most central features of the episodic memory feature network when they imagine the future. This pilot-study is the first to suggest what the pool of features people draw on may look like and the first to investigate what the relationships and structures within that pool are. In doing so, our pilot study identified directions for future research both for the method developed here and research on memory and future imagination more broadly. Most
constructed and how they are connected. All in all, this pilot study gives confidence that this is a line of research worth perusing on a larger scale.
References Auxiliary verbs. (2018) Retrieved
fromhttp://www.myenglishpages.com/site_php_files/grammar-lesson-auxiliary-verbs.php
Barrat, A., Barthelemy, M., & Vespignani, A. (2007). The Architecture of Complex Weighted Networks: Measurements and Models. Large Scale Structure and Dynamics Of
Complex Networks: From Information Technology to Finance and Natural Science, 67-92
Braver, T. S. (2012). The variable nature of cognitive control: a dual mechanism framework. Trends Cogn. Sci. 16, 106–113. doi: 10.1016/j.tics.2011.12.010
Burgess, P. W. & Shallice, T. (1996) Confabulation and the control of recollection. Memory, 4, 359-411. Doi: 10.1080/ 096582196388906
Cabeza, R., Rao, S. M., Wagner, A. D., Mayer, A. R. & Schacter, D. L. (2001) Can medial temporal lobe regions distinguish true from false? An event-related fMRI study of veridical and illusory recognition memory. Proc. Nati Acad. Sei. USA 98, 4805-4810. Doi:10.1073/pnas. 081082698
Conway, M. A., Singer, J. A., & Tagini, A. (2004). The self and episodic memory: Correspondence and coherence. Social cognition, 22, 491-529.
D'Argembeau, A., Stawarczyk, D., Majerus, S., Collette, F., Van der Linden, M., Feyers, D., ... & Salmon, E. (2010). The neural basis of personal goal processing when
envisioning future events. Journal of cognitive neuroscience, 22(8), 1701-1713. Doi: https://doi.org/10.1162/jocn.2009.21314
Dudai, Y., & Carruthers, M. (2005). The Janus face of mnemosyne. Nature, 434(7033), 567. Doi:10.1038/434567a
Hassabis, D., & Maguire, E. A. (2007). Deconstructing episodic memory with construction. Trends in cognitive sciences, 11(7), 299-306.
Lists of Adjectives. (2018). Retrieved
fromhttps://www.paperrater.com/page/lists-of- adjectives
Lists of Pronouns in English. (2018). Retrieved
fromhttp://www.really-learn-english.com/list-of-pronouns.html
Newman, M. E. (2001). Scientific collaboration networks. II. Shortest paths, weighted networks, and centrality. Physical review E, 64(1).
Painter, J. M., & Kring, A. M. (2015). Back to the future: Similarities and differences in emotional memories and future imaginations. Applied Cognitive Psychology, 29(2), 271-279.
Qualtrics, I. (2013). Qualtrics.com. Provo, UT, USA
R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.
Rosenbaum, R. S., Köhler, S., Schacter, D. L., Moscovitch, M., Westmacott, R., Black, S. E., & Tulving, E. (2005). The case of KC: contributions of a memory-impaired person to memory theory. Neuropsychologia, 43(7), 989-1021.
Schacter, D, L., & Addis, D, R. (2007). Constructive Memory: The Ghost of Past and Future. Nature, 445, 27. Doi: 10.1038/445027a
Schacter, D. L., & Addis, D. R. (2007). The cognitive neuroscience of constructive memory: remembering the past and imagining the future. Philosophical Transactions of the Royal Society B: Biological Sciences, 362(1481), 773-786.
Schnider, A. (2003) Spontaneous confabulation and the adaptation of thought to ongoing reality. Nat. Rev. Neurosci. 4, 662-671. Doi:10.1038/nrnll79
Szpunar, K. K., & McDermott, K. B. (2008). Episodic future thought and its relation to remembering: Evidence from ratings of subjective experience. Consciousness and cognition, 17(1), 330-334. Doi: https://doi.org/10.1016/j.concog.2007.04.006 Tulving, E. (1985). Memory and Consciousness. Canadian Psychology, 26, 1-12. Doi:
http://dx.doi.org/10.1037/h0080017
Tulving, E. (2002). Episodic memory: From mind to brain. Annual review of
psychology, 53(1), 1-25. Doi: http://doi.org/10.1146/annurev.psych.53.100901.135114 Tulving, E. (2005). Episodic memory and autonoesis: Uniquely human. The missing link in
cognition: Origins of self-reflective consciousness, 3-56.
Tulving, E., Schacter, D. L., McLachlan R., Moscovitch, M. (1988). Priming of Semantic Episodic Knowledge: A Case Study of Retrograde Amnesia. Brain and Cognition, 8, 3-20. Doi: https://doi.org/10.1016/0278-2626(88)90035-8
Wagner, A. D., Shannon, B. J., Kahn, I., & Buckner, R. L. (2005). Parietal lobe contributions to episodic memory retrieval. Trends in cognitive sciences, 9(9), 445-453. Doi:
https://doi.org/10.1016/j.tics.2005.07.001
Yasuda, K., Osamu, W., Yoshiharu, O. (1997). Dissociation Between Semantic and Episodic Memory. Elsevier, 33, 623-638. Doi: https://doi.org/10.1016/S0010-9452(08)70721-4
Zheng, H., Luo, J., & Yu, R. (2014). From memory to future imagination: what are the overlapping and the distinct components between remembering and
imagining? Frontiers in psychology, 5, 856. Doi:
