Part 1:Information Stored in Memory Afects Abductive Reasoning
Mar 19, 2022
for more information:ali.ma@wecistanche.com
Anja Klichowicz1 · Daniela Eileen Lippoldt1 · Agnes Rosner2 · Josef F. Krems1
Received: 26 March 2020 / Accepted: 7 December 2020 / Published online: 11 January 2021
© The Author(s) 2021
Click to Cistanche para que sirve for memory
Abstract
Abductive reasoning describes the process of deriving an explanation from given observations. The theory of abductive reasoning (TAR; Johnson and Krems, Cognitive Science 25:903–939, 2001) assumes that when information is presented sequentially, new information is integrated into a mental representation, a situation model, the central data structure on which all reasoning processes are based. Because working memory capacity is limited, the question arises how reasoning might change with the amount of information that has to be processed in memory. Thus, we conducted an experiment (N = 34) in which we manipulated whether previous observation information and previously found explanations had to be retrieved from memory or were still visually present. Our results provide evidence that people experience differences in task difficulty when more information has to be retrieved from memory. This is also evident in changes in the mental representation as refected by eye-tracking measures. However, no differences are found between groups in the reasoning outcome. These findings suggest that individuals construct their situation model from both information in memory as well as external memory stores. The complexity of the model depends on the task: when memory demands are high, the only relevant information is included. With this compensation strategy, people are able to achieve similar reasoning outcomes even when faced with tasks that are more difficult. This implies that people are able to adapt their strategy to the task in order to keep their reasoning successful.
Introduction
Inferring an explanation from a set of observations is one of the most challenging tasks our minds engage in every day. This process is called abductive reasoning (Johnson & Krems, 2001; Peirce, 1931) and is understood as one out of three classes of inference (abduction, deduction, induction, see Table 1, for an overview see Peirce, 1931). In deduction a rule (If E → O) and the explanation (E) is present, and the data (or observation; O) have to be inferred. In induction the explanation, as well as data (observations), is present and one has to infer the rule. In contrast, in abduction, an explanation is derived from observations given a rule (Josephson & Joseph- son, 1996; Meder & Mayrhofer, 2017; Peng & Reggia, 1990). It has been shown that fnding the explanation can further understanding (Lombrozo, 2006), it facilitates learning (Murphy & Allopenna, 1994; Williams & Lombrozo, 2010), can influence our judgments in terms of the perceived typicality of category members (Ahn, Marsh, Luhmann, & Lee, 2002; Murphy & Allopenna, 1994), and foster conceptual coherence (Murphy & Medin, 1985; Patalano, Chin-Parker, & Ross, 2006). Having explanations available puts us in a better position to predict and control the future (Lombrozo, 2006). Therefore, understanding how people infer that one explanation is more likely than another is central in understanding human thinking.
Abduction is a highly complex process because observations can lead to a combination of different explanations, but a single explanation can also account for a number of observations (Johnson & Krems, 2001). The best explana- tion is defned as the explanation with the lowest level of complexity. For example, when physicians attempt to infer a diagnosis for a number of symptoms, they have to integrate all the available information, some of which is not currently visible but may have to be recalled from patient reports, examinations, or laboratory tests.
Research on eye movements has found that, when retrieving information that was previously encoded at a specific spatial position, a person’s gaze, and respectively, the focus of attention, returns to this spatial location as an aid to working memory (Johansson & Johansson, 2014, 2020; Scholz, Klichowicz, & Krems, 2018). To remember all the relevant information, a physician might, therefore, look at
Table 1 Overview over the abduction, deduction, induction based on a rule (If E → O), an observation (O), and an explanation (E)
t
the patient’s fle without even opening it, using eye movements to facilitate memory retrieval. This example shows that not only online reasoning skills but also memory plays an important role in abductive reasoning, and that eye movements are used to access memory contents, even if no information is visible in the visual array (for overviews see Ferreira, Apel, & Henderson, 2008; Richardson, Altmann, Spivey, & Hoover, 2009; Wynn, Shen, & Ryan, 2019). It also shows the complexity of the task, as a high number of symptoms can result in different combinations of diagnoses. A number of process models have been developed to describe the process of abductive reasoning (e.g., TAR:
Johnson & Krems, 2001; HyGene: Thomas, Dougherty, Sprenger, & Harbison, 2008). In this study, we focus on the Theory of Abductive Reasoning (Johnson & Krems, 2001), which describes the process of abductive reasoning compre- hensively and not just parts thereof.
The reasoning process
TAR (Johnson & Krems, 2001) assumes that during sequential information presentation, new observations are integrated into a mental representation called a situation model. New information is thus assigned to slots in memory which, taken together, form an overall understanding of the current situation. Research has found evidence that this situation model is located in working memory (Böhm & Mehlhorn, 2008; Thomas, Dougherty, Sprenger, & Harbison, 2008). TAR (Johnson & Krems, 2001) calls this step to comprehend. In a previous study, we found that this situation model expands over the sequential presentation of observations, becoming more complex over the course of a reasoning task (Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020). That is, as more information has to be integrated, the number of spatial areas that are associated to explanations grows. As people gaze back at locations that contained information that has to be retrieved, the number of spatial areas looked at grows as well.
To form a correct situation model, is it necessary to have all previous information active in memory, according to research showing that the generation of explanations is influenced by the knowledge currently activated in memory (Mehlhorn, Taatgen, Lebiere, & Krems, 2011; Rebitschek, Krems, & Jahn, 2016; Thomas, Dougherty, Sprenger, & Harbison, 2008). Based on the situation model, the reasoner forms explanations of the observation. If able to form concrete explanations, the reasoner executes a consistency check in terms of the implications of these explanations for the situation model. If a combination of explanations can explain all observations without any discrepancy or redundancy, the process is successful. Tracking participant's eye movements during abductive reasoning indeed showed that observations, as well as explanations, must be part of the situation model as participants look at both locations during the retrieval of these information to make an inference (Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020).
However, our results revealed that explanations receive much more attention than observations. This can be explained by the fact that explanations may be subject to change throughout the reasoning process, whereas an obser- vation does not change once it has been made. This also shows that explanations that have already been found have more weight in the overall explanation of a set of observations. Following this argument, we assume that participants put more efort into keeping information that is highly relevant for the final explanation active. Observations might lose their relevance as soon as they are concretely explained, that is, their activation might decline.
Other research by Bauman et al. has shown that there are diferent levels of activation in working memory during abductive reasoning (Baumann, Mehlhorn, & Bocklisch, 2007). They found that explanations that are relevant for explaining current observations are kept in a more active state than irrelevant explanations. That is, information in the situation model is more active the more relevant it is to the process of reasoning. This is also due to limited working memory capacity (Baddeley & Hitch, 1974; Johnson-Laird, Byrne, & Schaeken, 1992).
To integrate new information into the situation model, one has to retrieve information that is already contained in the model. As the retrieval of information absorbs resources (Hayhoe, Bensinger, & Ballard, 1998), people only engage in active memorization and retrieval when necessary. This is also illustrated by a study by Ballard, Hayhoe, and Pelz (1995), who asked their participants to copy a pattern of colored blocks. They found that participants kept the requirements for memory as low as possible by using more eye movements to gather information from the environment when needed. In their study, participants never used memorization and retrieval as a strategy.
The aforementioned research suggests that a task is per- ceived as more difcult when the demands on working memory are high. This should also have an impact on the outcome of the reasoning process, as only information that is repre- sented can be taken into account in seeking an explanation. As stated earlier, the situation model as proposed by TAR (Johnson & Krems, 2001) is highly dependent on information that is active in memory (Mehlhorn, Taatgen, Lebiere, & Krems,
2011). On the other hand, information that is present in the outside world eliminates the need of keeping the information in mind (Gray & Fu, 2001). This raises the question whether the process of abductive reasoning changes if less relevant information does not decay because it can be stored in external memory (Gray & Fu, 2001; O’Regan, 1992) and, therefore, remains active without using working memory capacity.
Previous research suggests that people consider present information frst (O’Regan, 1992), for instance, the information presented on a computer screen. The total efort to handle information is a sum of the efort put into motor action (such as eye movements) and storage and retrieval in memory (Gray & Fu, 2001). As mere eye movements in a limited visual array do not pose high requirements to the motor system, we assume that people prefer getting information using small eye movements from the outside world rather than retrieving it (Gray & Boehm-Davis, 2000; Gray & Fu, 2001). People should, there- fore, be able to use and integrate more information when they do not have to retrieve it (Ballard, Hayhoe, Pook, & Rao, 1997; O’Regan, 1992; Spivey & Dale, 2011). Integrating information means that all previous information and more impor- tant explanations are taken together to fnd the least complex explanation for all observations. In contrast, when demands exceed working memory capacity, people tend to explain each observation separately without taking previous information into account. However, this results in a more complex overall explanation (Johnson & Krems, 2001). Further, as the situation model grows with each new piece of information and retrieval requires more resources than inferences from givens, it should also take more time to reconstruct a complex situation model that includes a number of observations from memory rather than from an external memory store.
Taking all these findings together, we can say that the situation model (a) is stored in working memory (b) as a result of limited capacity, (c) is task-dependent, (d) is crucial to the outcome of reasoning, and (e) might influence the processes of reasoning as it determines the amount of information considered.
Visual attention
The first point above (a), that the situation model is most likely held in working memory, is particularly important when investigating the content and structure of the situation model to make more elaborate assumptions regarding the information used in the reasoning process (e). As we know from a large body of literature, working memory is closely connected with visual attention, which is often reflected in eye movements (Belopolsky & Theeuwes, 2009; Huettig, Olivers, & Hartsuiker, 2010; Theeuwes, Belopolsky, & Olivers, 2009). As attention precedes eye movements (Deubel & Schnei- der, 1996) and, therefore, determines what we look at next, it influences what is stored in working memory (Theeuwes,Belopolsky, & Olivers, 2009). Also, within the mental representation that is held in working memory, shifts of attention occur (Grifn & Nobre, 2003; Theeuwes, Kramer, & Irwin, 2011) and function as a mechanism to rehearse and maintain information (e.g., Godijn & Theeuwes, 2012).
In essence, attention determines what is part of the mental representation. As one of the key functions of attention is orienting in visual stimuli (Posner, 1994), the stimulus and its complexity should also affect what is part of the mental representation. The amount of present information should not only affect where attention is guided, but also the amount of information that is integrated into the situation model, as the task determines what is processed within and across gaze positions (Hayhoe, Bensinger, & Ballard, 1998). Taken together, manipulations of the task with regard to the amount of given information influence the reasoning process as it guides attention, which in turn determines what information enters the mental representation in working memory. As attentional shifts manifest themselves in eye movements, gaze data are able to shed light on the question of what information is used as we engage in abductive reasoning.
Research objectives
Following the study of Ballard, Hayhoe, and Pelz (1995), we expect that people experience more difficulties when information has to be retrieved from memory than when information is given or can be derived from givens. As a consequence, study participants should experience the task as more demanding. Thereby, the workload was operationalized with eye-tracking as behavioral data. We assume that retrieval requires more cognitive resources than inference or gathering informa- tion from a visual setup. As eye movements to empty information locations are more enhanced when memory demands are high (Kumcu & Thompson, 2018; Scholz, Mehlhorn, Bock- lisch, & Krems, 2011), we expect retrieval to result in more pronounced eye movements to information locations.
Further, we expect that study participants use all information provided when seeking an explanation for a set of observations. However, when information has to be retrieved from memory, we expect participants to focus primarily on the information that is most important to the task as retrieval absorbs more resources. This should also have an impact on the outcome of the reasoning process.
In order to grasp differences in abductive reasoning based on the amount of given compared to retrieved information, we investigate three questions:
(1) Do participants experience differences in the difficulty of the task based on the amount of currently given information?
(2) Does the process of reasoning change when more information is given? To be precise is the number of items
Fig. 1 The black box during the last observation of the same trial in the different conditions. In condition A&O, all atoms and observation locations of the former three observations are still visible. Only previously placed atoms remained visible in condition A. In condition O only all previous observation locations were visible. The current observation and the corresponding atom but nothing more remained visible in condition N
integrated into the situation model smaller when those items have to be retrieved from memory in comparison to when these items are present in the visual array?
(3) Does the reasoning outcome change because people use more information for an explanation when they do not have to retrieve it?
This study
To investigate outcomes as well as the reasoning process depending on the given or retrieved information, we used the same visuospatial reasoning task as in our previous study (Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020). The“black box task”(BBX; Johnson & Krems, 2001; Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020) is a tool that can be used to study the abductive reasoning process in detail. In this task, a box is presented to study participants, who are asked to infer what objects are inside it by interacting with it. The participant’s precise task is to locate a number of atoms hidden by inferring the path of light rays from their observed entry and exit positions. The entry and exit positions of the light rays represent the observations and are fixed and for all participants similar. The path develops as the light rays interact with the hidden atoms. How this interaction takes place is defined by a small number of rules. Assumptions regarding the locations of the hid- den atoms represent the explanations. As explanations must be inferred from rules, we are able to trace the generation of causal explanations rather than the retrieval of learned associations or past instances stored in memory (e.g., Klahr & Dunbar, 1988; Thomas, Dougherty, Sprenger, & Harbison, 2008). In other words, the task allows us to investigate how the situation model evolves. Over the course of the experiment, we manipulate how many of the observations and explanations remain visible within the visual field of the participants and how much of the information gathered has to be stored in memory.
In this experiment, we introduced four conditions, which we manipulated in a within-subject design. Throughout the trial, all atoms and observations (condition A&O), only atoms (condition A), only observations (condition O), or neither atoms nor observations (condition N) remained visible in the black box display (see Fig. 1). As stated earlier, we were interested in how this affects response behavior as well as the process of abductive reasoning itself. The manipulation sheds light on the question of how the situation model changes depending on the amount of information that has to be stored in memory.
Using eye movements as a method to assess memory retrieval
Eye movements have long been known as a good process measure (Hannula, Althof, Warren, Riggs, Cohen, & Ryan, 2010; Holmqvist et al., 2011) for processes with information from visually presented givens. Ballard, Hayhoe, Pook, and Rao (1997), for instance, report higher sensitivity of eye movement measures than of conscious reports by study participants. Further, eye movements are tightly coupled with retrieval processes from memory (e.g., Scholz, Mehlhorn, Bocklisch, & Krems, 2011; Scholz, Mehlhorn, & Krems, 2016). For instance, they play a role in constructing and maintaining the mental image (Brandt & Stark, 1997; Laeng, Bloem, D’Ascenzo, & Tommasi, 2014). Therefore, eye movements are a valuable means of investigating both given and retrieved information. Given information results in spatial indexing, whereas information that has to be retrieved from memory elicits memory indexing. The memory indexing method (Jahn & Braatz, 2014; Renkewitz & Jahn, 2012) utilizes the fact that, when retrieving information, people’s gazes are drawn to the spatial location where the information was previously encoded, even if that information is no longer displayed (see looking-at-nothing phenomenon; Laeng & Teodorescu, 2002; Richardson & Kirkham, 2004; Scholz, Klichowicz, & Krems, 2018; Spivey & Geng, 2001, for an overview see Ferreira, Apel, & Henderson, 2008; Richardson, Altmann, Spivey, & Hoover, 2009). The encoded information is thus stored along with a spatial index (e.g., Kumcu & Thompson, 2016; Pylyshyn, 2001; Richardson & Kirkham, 2004). Probing this information reactivates the associated spatial index, which elicits eye movements toward the (now empty) spatial location.
As visual processing is task-driven (Hayhoe, Bensinger, & Ballard, 1998; O’Regan, 1992; O’Regan & Levy-Schoen, 1983), the amount of given information during a task might result in different eye movement patterns. In the study by Ballard, Hayhoe, and Pelz (1995), displays changes affected study participants' eye movements depending on where they were in the task at that moment, suggesting that vision only considers features that are currently task-relevant.
Hypothesis 1: differences experienced in task difficulty
People are generally able to engage in abductive reasoning successfully. However, the question remains whether they experience differences in task difficulty even when they are successful. To compare experiences with actual outcomes, we introduce a subjective rating of difficulty of each condition. Following Ballard, Hayhoe, and Pelz (1995), we assume that retrieval poses more demands on participants than acquiring information from the visual setup. This should be evident from the participant's ratings of the difficulty of conditions. Conditions with visible atom and observation locations (A&O) should be experienced as easiest. We have no assumptions regarding differences depending on whether participants see former explanation (atom) locations (condition A) or can reconstruct them based on observation locations (condition O). However, as retrieval is assumed to be a more demanding process, we expect study participants to rate the condition in which they have to remember observation as well as explanation locations (N) as the most difficult.
Hypothesis 2: elements of the situation model
As previous explanations are more important for the situation model than previous observations, we predict that study participants spend more time on previous explanation locations than on previous observation locations, regardless of whether explanations or observations are still visible on the screen. Therefore, we assume that participants' visual attention is driven to previous explanation locations irrespective of condition. If previous explanations are still visible, participants look at them to use the external memory store for the construction of a coherent situation model. If explanation locations are not visible, participants still look at their location as they either construct (if observations are still present) or retrieve (if no previous information is visible) their position in assessing the situation model. As the degree of activation is much smaller for observation locations (Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020), we assume that participants only look at them when present. Otherwise, the costs of retrieval are too high, as only explanations are relevant for the overall explanation. We expect that participants look at observation locations in order to infer previous explanation locations if previous explanations are not visible.
Hypotheses 3: integrative solutions
As the external memory store can act as an aid to relieve working memory, Hypothesis 3a proposes that more information is considered to fnd the best explanation for all observations when information remains visible on the screen. That is, a less complex explanation is used. We call this explanation“integrative” as it integrates previous explanations in order to explain new observations and does not use a new explanation for every observation. Hypothesis 3a, therefore, states that study participants find more explanations that integrate information to a higher degree when more information remains visible and when the setup acts as an external memory store. It is, therefore, irrelevant whether observation locations or explanation locations are still visible. Even though explanation locations are more important for the overall explanation, observation locations can be used to infer explanations rather than to retrieve them.
As a setup that requires keeping all information in memory requires participants to construct, maintain, and retrieve the situation model as needed, we propose that participants take more time to find a coherent explanation in this case. To be more precise, we hypothesize that finding the explanation for the last observation takes more time when atoms and observations have to be retrieved than when information is given (Hypothesis 3b).
Method
Participants
For 34 participants, the calibration of the eye tracker succeeded to an accuracy of at least 2° of visual angle. Due to decreasing eye-tracking accuracy throughout the experiment, three participants had to be excluded. The remaining 31 participants (17 females, 14 males) were all students from Chemnitz University of Technology and had a mean age of 22.7 years (SD = 3.7). All had normal or corrected to normal vision.
Apparatus
The task was presented on a 22″ computer screen (1680 × 1050 pixels), which was located at a distance of
Fig. 2 Rules of the black box task (BBX). A light ray entering the black box can take the following paths, depending on where the ray hits the hidden atom: 1 = straight through, 2 = L-pattern, 3 = absorption, 4 = U-pattern, 5 = Z-pattern
63 cm in front of the participants. E-Prime 2.0 was used to present stimuli and participants responded with a standard keyboard and via mouse. At a rate of 120 Hz, an SMI RED remote eye-tracking system sampled data from the right eye during the reasoning task. We used iView X 2.5 to record data following five-point calibration, and BeGaze 3.0 to analyze gaze data with a fixation dispersion threshold of 100 pixels and a duration threshold of 80 ms. Further, we used IBM SPSS statistics 24, Microsoft Excel 2016, JASP 0.8.4.0, and R version 3.4.3 to conduct the analysis.
The black box task
The black box task (BBX) was defined by a 10 × 10 grid with a size of 25.92° × 26.27° of visual angle (1015 × 1029 pixels). In this grid, the participant's task was to locate hid- den atoms by following where light rays entered and exited the box. The actual path of the light ray through the box remained hidden from sight. Participants only saw the predefined entrance and exit positions of each ray as indicated by a number appearing at the border of the black box (see Fig. 2).1
These entry/exit locations are observations according to TAR (Johnson & Krems, 2001) and require one or more explanations, which were operationalized as atoms. Each atom is surrounded by a field of influence (a circle around the atom). As shown in Fig. 2, hitting this field, ray and atom interact according to a set of predefined rules that the participants learned beforehand. Following the numbers that indicated the observation positions of each ray, participants
could place atoms using the mouse. Even though participants did not have to place atoms, they were instructed to place each new explanation as early as possible. It was up to participants to decide when to move on to the next trial by pressing the space bar. A digit in the upper left corner of the black box (see Fig. 1) indicated the number of observations left during one trial.
The black box task allowed five different rules, which were as follows: when the ray does not meet any field of influence of an atom, it finds its way straight through the black box. An L-pattern arises if the ray hits the field of influence at an angle and is reflected 90°. A ray of light is absorbed and does not exit the black box if it hits an atom directly in the middle. Combinations of two L-patterns can result in a U- or Z-pattern.
Straight through, L-pattern, absorption, U-pattern, and Z-pattern were the names used to describe the observations for a better understanding of the task. However, the actual observations only consisted of entry and exit locations of the ray. The names of the observed patterns describe the most likely path the ray of light would take through the black box based on the observation locations.
After each presentation of a new observation location, participants were asked to infer and place the atom based on the rules explained above. Participants were instructed to keep the number of atoms to explain a ray pattern as low as possible throughout the trial. Each trial consisted of four observations in sequential order, which were indicated by a number at the entry and exit position of the ray in the BBX (Fig. 2).
All participants solved 12 trials in each of the four conditions with differing amounts of information that had to be stored in memory. In the first condition, all atoms and observation locations remained visible throughout the trial. All information could, therefore, be placed in an external memory store. The conditions were named based on the items that remained visible. Since atoms and observation locations remained in the first condition, it was referred to as A&O. In a second condition, only already placed atoms remained visible (condition A). During the 12 trials of the third condition, only observation locations of the rays remained visible and atom locations had to be remembered (condition O). Because the last condition was completely memory-based and “nothing” remained in the external memory store, it was referred to as N for nothing. All trials consisted of four observations; therefore, Fig. 1 shows what was presented during observation four in each condition. Even though we used the same example for better understanding in Fig. 1, participants did not solve the same trials in each condition, but slight variations that were balanced in complexity and difficulty to prevent learning effects.
Even though participants were instructed to include a preferably small number of atoms in the final explanation
of the trial, two-thirds of the trials could be solved in two diferent ways. In the following, these trials are called experimental trials. First, participants could explain each observation separately without considering previous atoms (Fig. 3a). Second, participants could keep the number of atoms low by using previously set atoms to explain the last observation (Fig. 3b). Because this means that all other atoms had to be integrated to find the explanation, we call the trials that were solved “integrative”. The remaining third of trials during the test phase are called distractor trials and had only one correct solution.
