What Can We Learn by Treating Perspective Taking as Problem Solving?

Abstract

Perspective taking has been studied extensively using a wide variety of experimental tasks. The theoretical constructs that are used to develop these tasks and interpret the results obtained from them, most notably theory of mind (ToM), have conceptual shortcomings from a behavior-analytic perspective. The behavioral approach to conceptualizing and studying this class of behavior is parsimonious and pragmatic, but the body of relevant research is currently small. The prominent relational frame theory (RFT) approach to derived perspective taking asserts that “deictic framing” is a core component of this class of behavior, but this proposal also appears to be conceptually problematic. We suggest that in many cases perspective taking is problem solving; when successful, both classes of behavior involve the emission of context-appropriate precurrent behavior that facilitates the appropriate response (i.e., the “solution”). Conceptualizing perspective taking in this way appears to have many advantages, which we explore herein.

Perspective taking is an ability of critical importance that helps us establish and maintain relationships, negotiate deals, predict the actions of others, and achieve a wide variety of other valuable outcomes. This ability has been studied extensively by psychologists, who have approached the topic using many different theoretical constructs and experimental procedures. In this work, we present some common theoretical and experimental approaches to perspective taking, followed by a description of a behavioral approach to perspective taking and problem solving. In brief, problem solving relies on the emission of overt or covert behaviors that produce supplementary stimuli that in turn evoke behavior that may produce the reinforcer (i.e., a solution to the problem). Problem-solving behavior is evoked in contexts in which a discriminative stimulus (S^D) associated with a reinforcer is present, but the response that is required to obtain the reinforcer cannot be emitted directly. Because much perspective-taking behavior, including behavior in experiments designed to study this class of behavior, appear to qualify as problem solving, we make the case that behavioral theory that has been developed for the purpose of understanding problem solving can serve as a strong foundation for understanding perspective taking. A variety of assessments and tasks are used to study perspective taking. These can be roughly divided into two categories: those focusing on visual perspective taking (VPT), and those focusing on mental state attribution, typically referred to as theory of mind (ToM).

Visual Perspective Taking

VPT tasks include spatial components, such as the current position of a subject (viewer), the target (an experimenter, a doll, or another person), and the position of other objects in relation to the subject and the target. VPT is often subdivided into Level 1 and Level 2 (Flavell et al., 1980, Flavell et al., 1981; Flavell, 2004; Moll & Meltzoff, 2011). At Level 1, a subject demonstrates the ability to differentiate the perspectives of self and others based on what can be seen by each party. For example, a two-sided card with a different picture on each side is held up between a child and a teacher, and the child is asked, “What can you see, and what can I see?” Children are also assessed for their VPT Level 1 ability using tasks based on line-of-sight paradigms. In one relevant experimental protocol, researchers placed a doll on one side of a cardboard screen and a desk on the same side as the doll, or the opposite side, and asked children to indicate whether the doll could see the desk (Leslie & Frith, 1988). In another relevant study, Baron-Cohen (1989) examined whether preschool children could identify which object an experimenter was looking at, when only the experimenter’s gaze (not head orientation) was available as a cue.

Although VPT Level 1 tasks are meant to reveal the participant’s ability to identify if others’ perspectives differ, VPT Level 2 tasks are meant to reveal the participant’s ability to identify how others’ perspectives differ from their own. In one relevant task, the researcher places a card with a picture flat on a table between the participant and the researcher. The participant is then asked, for example, “When I look at the picture, is the turtle the right way up or upside down?” (Flavell et al., 1981, p.101). The three-mountain task is a VPT Level 2 task that assesses the participant’s ability to determine how perspectives differ based on a vantage point in a scene based on selection of pictures that correspond with specific vantage points. The reality–appearance distinction task is also said to provide information about Level 2 perspective taking. In one such task, the researcher gives a child a sponge that appears to be a rock. The child is first asked what the object looks like and then asked, “is it really and truly a sponge or is it really and truly a rock?” The child’s ability to discriminate between what an object appears to be at first sight and what it really is (e.g., a sponge that looks like a rock or a candle that looks like a pencil) is considered evidence of VPT Level 2 ability, which is correlated with the ability to distinguish between appearance for the self and appearance for another person (Flavell, 1986; Starr & Baine, 1996; Moll & Meltzoff, 2011).

VPT is also evaluated using spatial tasks that measure one’s understanding of how objects in the environment are oriented in relation to one another (e.g., behind, below, to the right of). From one’s own perspective, this type of task is relatively straightforward, but tasks involving identification of the relative positions of objects from another perspective are less so. One potential strategy for solving such tasks is “egocentric mental rotation,” which involves imagining the perspective of another person or point of view (Kessler & Thomson, 2010; Michelon & Zacks, 2006; Surtees et al., 2013). Object rotation is another possible strategy that involves imagining the movement of an object relative to the viewer and the environment. A variety of studies have been conducted to understand the conditions under which participants engage in object rotation or egocentric mental rotation when completing this type of VPT task. For example, Michelon and Zacks (2006) asked participants to identify whether an object was to the left or to the right of a doll. They found that the response latency increased with an increase in the angular distance between the participant’s and the doll’s orientation, which they interpreted as evidence for participants applying egocentric mental rotation under these conditions.

Theory of Mind

The second general category of perspective taking, typically referred to as theory of mind (ToM), emphasizes the role of mental attribution in perspective taking. Originally proposed by Premack and Woodruff (1978), ToM refers to one’s ability to differentiate their own perspective from the perspectives of others by making inferences about others' informational states (i.e., thoughts and feelings). This general ability is also sometimes referred to as “mind reading” (Howlin et al., 1999). A broad variety of tasks have been developed with the intent of revealing participants’ ToM capabilities.

Unexpected transfer or identity tasks are meant to reveal one’s ability to identify “false beliefs” held by another individual. The most well-known ToM task of this variety is the Sally–Anne test (Baron-Cohen et al., 1985), which was introduced by Wimmer and Perner (1983). In this false-belief task, a child is introduced to a vignette, such as, “Sally and Anne were in the room. Sally was playing with her marble, then placed it in her basket and left the room. Anne then put it into her basket. Later, Sally comes back to the room to play with her marble.” The child is asked questions such as, “Where does Sally think her marble is?” to see if the child can identify the false belief held by Sally. There are many variations of this test. The unexpected identity task is another test developed to measure a child’s ability to identify a false belief. In the task, an examiner shows the participant a container for an object (like a sweets packet), but the container actually holds another object (e.g., a pencil). After the participant opens the container and acknowledges what is inside (e.g., they say that they see a pencil), a doll is introduced to the child. The child is given a prompt, such as, “Here comes Rosie. What will she think is inside?” The unexpected identity task is similar to the appearance–reality test because both assess a child’s ability to distinguish two identities of the same object (e.g., a sponge that looks like a rock or a candy box that contains pencils), but the unexpected identity task focuses on another person’s false belief prior to their exploration of the object.

Picture sequencing is used to test a participant’s understanding of typical causal outcomes, including those involving people engaging in various activities, and is meant to provide an indication of a participant’s ToM capability (Baron-Cohen et al., 1986). Some sequences focus on interactions among people (e.g., a child crying over his ice cream being snatched) or violations of someone’s expectations (e.g., someone being surprised about a missing item). The experimenter provides the first picture, then asks the participant to look at the other pictures and arrange them into the correct sequence; the participant is then asked to narrate the story.

The “reading the mind in the eyes test” was developed as a measure of “advanced” ToM ability in high-functioning adults with autism or Asperger syndrome (Baron-Cohen et al., 1997; Baron-Cohen et al., 2001). This test requires a subject to match a term to a picture of eyes without any other contextual information provided. For example, the subject selects a correct term (e.g., “serious”) from the other choices, such as “ashamed,” “alarmed,” and “bewildered.” Another measure of advanced ToM abilities known as the “faux-pas test” assesses the ability of both children and adults to identify socially awkward behavior using 20 short stories, each containing a faux pas incident (Baron-Cohen et al., 1999; Stone et al., 1998). A similar task is also used to measure one’s comprehension of others’ mental states in different situations, such as persuasion or a white lie, and to make inferences about the characters’ thoughts, feelings, and intentions (Happé, 1995). Hutchins et al. (2008) developed an assessment battery that combines various ToM tasks, the ToM task battery inventory (ToMI). It includes items assessing inferences about emotions and beliefs in addition to basic false-belief tasks such as the Sally–Anne test. Howlin et al. (1999) developed a teaching guideline for children with ASD to improve their “mind-reading” ability, combining different ToM tasks ranging in complexity from simple visualization tasks to more complex false-belief tasks used in the various ToM studies for children.

There are also interview-based assessment tools for measuring ToM. The “theory of mind assessment scale” (Th.o.m.a.s) is a qualitative tool developed especially for individuals with schizophrenia (Vallana et al., 2007). It consists of 39 open-ended questions investigating the interviewee’s understanding of their own and others’ perspectives (e.g., “Do you notice when others feel good?” “Do others notice when you feel good?”). The interpersonal reactivity index (IRI) questionnaire is used in many perspective-taking studies to measure attitudes in perspective-taking (Galinsky et al., 2008; Van der Graaff et al., 2014; Vilardaga et al., 2012; Barbero-Rubio et al., 2016; Kavanagh et al., 2018). It contains nine items for assessing the test taker’s “spontaneous” perspective-taking ability (e.g., “When I’m upset at someone, I usually try to ‘put myself in his shoes’ for a while”), indicated using a scale from “strongly disagree” to “strongly agree.” One issue with interview-based ToM measures is that they rely upon self-report and, therefore, outcomes may not correspond with behavioral measures of perspective taking.

The search for evidence of ToM extends to studies with nonhumans and infants as well, and tasks without an explicit language requirement have been developed for this purpose. In the anticipatory looking test, a subject's eye movement and duration are recorded by an eye tracker while the subject watches a series of short video clips showing actors (e.g., a human or a doll) performing false-belief tasks. In general, the experiment starts out with a “familiarization” procedure in which a subject is shown video footage demonstrating that a target object (e.g., a toy) is hidden in one of two locations. Then, the actor searches for it in its true location, where the target object was hidden. This is followed by a second video showing a false-belief component in which the object is relocated from one location to another while the actor is absent from the scene. When the actor returns to the scene, the initial direction and duration of the participant’s gaze are recorded and used to draw conclusions about where the participant predicts the actor will look for the toy (Krupenye et al., 2016, 2017; Onishi & Baillargeon, 2005; Senju et al., 2011; Southgate et al., 2007).

The Guesser-Knower task was also developed to reveal information about a participant’s ToM abilities. In this task, a participant sees two actors, a “knower” who has access to information about the location of a target (typically food) and a “guesser” who does not. For example, in a “guesser-absent” probe trial, the guesser leaves the experimental room, and the knower remains in the room while a baiter puts food into one of two food containers. The participant’s view of the containers, however, is occluded, so they cannot see where the food was placed. Upon the guesser’s return, the guesser and knower pose in the same manner but point and/or gaze at two different containers. The participant is then allowed to approach one of the containers. Approaching the container referenced by the “knower” is typically interpreted as evidence of ToM (Povinelli et al., 1990). Other versions of this task have been used, including tasks in which the guesser simply looks away while the food is being placed, and several species have been tested, including dogs (Catala et al., 2017; Maginnity & Grace, 2014) and scrub jays (Emery & Clayton, 2001). In a related task, researchers tested chimpanzees’ and human infants’ ability to predict the actions of others wearing opaque versus translucent goggles (Karg et al., 2015; Meltzoff, 2007; Meltzoff & Brooks, 2008; Senju et al., 2011; Vonk & Povinelli, 2011).

Another approach to testing ToM ability in nonhumans is to see whether they approach forbidden food (e.g., in situations where a human instructs a dog not to eat a piece of food on the ground). Various conditions can be set up to test an animal's performance. For example, Bräuer et al. (2004) compared dogs’ tendency to approach the forbidden food when their owner’s view was blocked by a barrier and when it was not. Other variations of this task have also been tested, including conditions in which the human observer is blindfolded or simply looking away rather than obstructed by a barrier (Bräuer et al., 2004; Kaminski et al., 2009). If the participant is more likely to take food when the observer’s view is obstructed, this is interpreted as evidence of ToM (Anderson et al., 1995; Anderson et al., 1996; Povinelli et al., 1996). In another type of study, called a begging task, if subjects approach and beg from a “seer” (whose view is not obstructed and who is capable of delivering food) more readily than a “blind” experimenter (whose view is obstructed or who is otherwise unable to respond to the participant’s begging), the participant is said to be demonstrating ToM (Udell et al., 2011).

Perspective Taking Controversy

There are many other tasks that have been used to study VPT and ToM. Our aim in providing descriptions of some of the commonly used tasks is to paint a picture of the conditions under which researchers consider behavior to be demonstrative of “perspective taking” or ToM. Although we have grouped these tasks according to the theoretical construct (VPT or ToM) that they are most closely associated with in the literature, there is some inconsistency in the literature regarding which construct a task is most relevant to. For example, the reality–appearance distinction task and the unexpected identity task are similar but are associated with two different constructs. There is also considerable disagreement about whether these two constructs have a “common cognitive denominator” (Moll & Meltzoff, 2011; Andrews et al., 2003; Andrews et al., 2012) or whether VPT and ToM are unique abilities. ToM tasks are typically more “conceptual,” often involving storytelling, story comprehension, and reporting of emotions (Aichhorn et al., 2006; Flavell, 1986; Gopnik & Astington, 1988), whereas VPT tasks are more commonly spatial tasks, but there are many exceptions to these general rules regarding the type of tasks used to study each construct (e.g., the guesser-knower task does not seem to fit the description of a typical ToM task). The face-value suitability of these tasks for studying each construct appears to be the sole consideration underlying the development of most perspective-taking tasks, which is problematic not only because constructs such as ToM are loosely defined, but also because the intentions of the task designer appear to heavily influence the interpretation of the outcomes obtained from the tasks.

These and other issues associated with ToM measures and the interpretation of study outcomes have been raised in the context of researching ToM in nonhumans. Penn and Povinelli (2007) critically evaluated the evidence that has been produced regarding ToM in nonhumans and concluded that none of the experimental protocols that have been used for this purpose are capable of distinguishing between “mind reading” (i.e., ToM) and “behavior reading” (i.e., stimulus control). Van der Vaart and Hemelrijk (2014) reviewed the evidence associated with this and other criticisms of research on ToM in nonhumans and concluded that Penn and Povinelli’s conclusion is “irrefutable.” In addition, van der Vaart and Hemelrijk found that many of the reviewed studies lacked appropriate control conditions and that the researchers frequently demonstrated a strong bias favoring ToM interpretations, even in the face of weak or conflicting evidence. For example, in studies employing the begging task with apes, van der Vaart and Hemelrijk found that researchers often emphasized evidence that supported ToM explanations and ignored evidence that did not. In studies examining recaching behavior in scrub jays and Clark’s nutcrackers when other birds are present, opposite outcomes related to the importance of line of sight between the observing bird and the cache were both interpreted as favoring a ToM explanation.

Some researchers have suggested that mind-reading-based interpretations of the outcomes of these studies are more parsimonious than behavior-reading explanations (Emery & Clayton, 2008; Tomasello & Call, 2006) but others have pointed out that predicting the actions of others based on representations of internal states in others is much more complex than predicting their actions based on observable stimuli (Barrett, 2010; Penn & Povinelli, 2007; Shettleworth, 2010; van der Vaart & Hemelrijk, 2014). Penn and Povinelli (2007) made some recommendations for protocols that might be used to evaluate ToM in nonhumans, including tasks for chimpanzees that are similar to the Sally–Anne task in which the location of a food item may be changed in the presence or absence of a dominant chimpanzee. The added complexity, however, does not eliminate the possibility for “behavior reading” explanations of success in these tasks. Van der Vaart and Hemelrijk (2014) were confident that adding such complexity would result in task failure, because chimpanzees regularly fail to “demonstrate ToM” in simpler tasks. Furthermore, failure in such a task would lead to no conclusive interpretations with respect to ToM. Penn and Povinelli rejected the notion that their criticisms apply to the protocols that have been used to study ToM in humans, but to support this point, they simply referred to the volume of ToM literature. Shettleworth (2010) and Barrett (2010) suggested that humans can intuit unobservable states in others but that we rarely do. That is, much of human perspective taking that is attributed to “mind reading” (i.e., ToM) is actually “behavior reading.”

We suggest that these critics are on the right track but that the murky nature of the “ToM” concept has complicated the task of exploring the mechanisms that give rise to specific examples of perspective taking and scientifically evaluating perspective taking in humans and other species in general. The topic of ToM is entangled with topics of concern that are fundamental to general psychology, including concerns about the subject matter of psychology itself and the role of folk psychology explanations in the field of psychology. Behavioral explanations for perspective taking are viewed by some psychologists as “killjoy” (Heyes, 1998; Shettleworth, 2010) explanations that lead to “despair” (Tomasello et al., 2003), but behavioral explanations are undoubtedly the starting point for a scientific understanding of perspective taking and must be exhausted before alternative explanations can be considered if we are to abide by the parsimony principle. Moreover, the behavioral approach to perspective taking is not limited to exploration of perspective taking behavior that can only be directly controlled by stimuli that are present in the immediate environment (i.e., “behavior reading). A wide range of behavioral theory and experimental tools have been developed to account for (and change) the type of behavior that is involved in “mind reading” or “ToM” explanations of perspective taking, including conversing, thinking, and imagining. The functional approach that characterizes behavioral psychology offers valuable insight into perspective taking that can be translated into effective applications. We now explore the behavioral approach to perspective taking before presenting our main hypothesis regarding the potential value of treating perspective taking as behavior analysts treat problem solving.

Perspective Taking: The Behavioral Approach

The behavior of participants in perspective-taking tasks and in everyday perspective-taking scenarios can be evaluated using the methods and principles of behavior analysis. This approach to understanding perspective taking is parsimonious and pragmatic, but it has yet to receive broader recognition among scientists and practitioners studying perspective taking. Behavioral interpretations of findings from VPT tasks are not necessarily at odds with general psychology interpretations, but interpretations that refer to poorly defined hypothetical constructs are not compatible with the behavioral approach. The most common such explanation for perspective-taking behavior, ToM, is a patent example of a loosely defined hypothetical construct. Schlinger (2009) described ToM as a “meme,” and Baum (1998) suggested that looking for ToM is no different from looking for the “soul.” Depending on how one defines the soul or ToM, it may be discovered, or not, in any individual. Moreover, the ToM concept leads to circular reasoning: observing certain behavior in an individual causes some researchers to conclude that they “have a ToM” and when asked why this behavior has been emitted they may explain that it is because they “have a ToM” (Schlinger, 2017). Of course, there is a definition attached to the label: ToM refers to an individual’s belief that another individual has their own thoughts, beliefs, and feelings (i.e., another perspective), but this does not lend any clarity to the concept. For instance, when searching for evidence that apes “have a ToM,” researchers are asking a question like, “does this ape believe that people or other apes have their own thoughts, beliefs, and feelings?” This begs the question, what does it mean to “believe” that another individual has their own private thoughts. Is this process not dependent upon advanced language capabilities (Gordon, 1998; Gray & Russell, 1998)? Even if it were possible to conclusively determine that an individual has ToM, because we cannot begin to understand what this means, how could this information benefit anyone?

In contrast, the behavioral approach to perspective taking involves, first, identifying the behaviors of interest (e.g., the behaviors that lead some researchers to conclude that an individual has a ToM); second, investigating the present stimulus conditions and the learning history that is responsible for the emission of such behavior. A social construct such as ToM cannot be the cause; the cause is to be found in the biology and the environment of the organism. Statements (whether overt or covert) about what others think and feel are not ultimate causes of perspective taking behavior, though they may play a part in the process of perspective taking. Instead, these behaviors are a type of perspective taking behavior that must also be explained. Evidence for perspective taking (including ToM) has come from a variety of tasks, some of which do not have any verbal1 component (especially those used for very young children and nonhumans) and others that are partially or entirely reliant on the verbal ability of the participant. We will briefly examine fundamental behavioral processes that must contribute to all perspective taking tasks and then review some higher order processes that are relevant to the causes of behavior in perspective taking tasks with a verbal component.

Perspective Taking: Fundamental Behavioral Processes

The key to understanding performance in any perspective taking task lies in an understanding of stimulus control. Unless the experimenter has presented an unsolvable problem, all of the stimuli that are required to evoke a correct response will be present in the task, assuming that the individual who is participating in the task has the appropriate biological constitution and learning history to respond appropriately to those stimuli. Failure to respond accurately in any perspective-taking task will necessarily be the result of a deficit in the stimuli that are present or a result of those stimuli not having the necessary functions. Of course, stimuli may be “present” but not accessed by the participant (e.g., they may have looked away during a critical stimulus change) and the functions of the stimuli in the task may be temporarily altered by motivating operations (Edwards et al., 2019a, 2019b).

Stimulus control can be simple, as when a discriminative stimulus evokes a response that has historically been reinforced in its presence. However, and this is particularly relevant to perspective-taking tasks, the evocative functions of discriminative stimuli can be conditional upon the presence of other stimuli. For example, an infant’s interaction with a specific toy is more likely to be reinforced (e.g., by joint play and attention from an adult) if the adult is looking at or pointing at the toy. As a result, the toy becomes a more effective evoker of relevant interaction when the adult’s gaze is directed toward the toy (Butterworth & Cochran, 1980; Butterworth & Jarrett, 1991; Gewirtz & Pelaez-Nogueras, 1992; Pelaez et al., 2012). Conditional stimulus control emerges from the conditional nature of relationships among events in the environment and learning mechanisms that are sensitive to such relationships. In experimental tasks designed to evaluate perspective-taking abilities, the correct response depends on which sequence of stimulus changes the participant has been exposed to. For example, in the guesser/knower task, identification of the “knower” is conditional upon the presence of stimuli that would result in them “knowing” (e.g., the placement of an object in the same direction as their gaze). In this well-studied task, we can be sure that there is no stimulus deficit. Therefore, if a participant fails the task, after addressing attentional and motivational factors, a behavioral researcher would conclude that the failure must be a result of deficiencies in the individual’s history of learning with those stimuli or, as a last resort, biological constraints on the individual’s ability to learn to respond appropriately to those stimuli.

A critical aspect of stimulus control that must be considered is stimulus generalization, the control of behavior by stimuli that are formally similar to but not the same as stimuli that were present when learning took place. The stimuli that control behavior in the present environment will necessarily differ from those that were present during conditioning in the past, so stimulus generalization is an essential feature of any learning mechanism. Stimulus generalization within a dimension (e.g., wavelength) can be quantified and, with additional training, stimulus generalization can be increased or decreased (Ghirlanda and Enquist, 2003). In the guesser/knower task, a successful participant will have had experience with people looking at objects being placed in occluded places and then going to those places to retrieve those objects, but the specific stimulus features of the people, places, and objects will have varied. Even though the specific stimuli and people used in the task differ from the many specific stimuli to which the individual has been exposed previously, they may still evoke the correct response. Failure to respond “correctly” in a guesser/knower task by a participant that appears to have an appropriate learning history could be a result of insufficient stimulus generalization.

Applying the parsimony principle, these fundamental and well-established behavioral processes must be considered before behavioral scientists consider other potential explanations for success or failure in perspective-taking tasks of all types. Failure to understand stimulus control mechanisms can lead to the erroneous conclusion that participants in a study do not demonstrate stimulus control when successfully completing a perspective-taking task. For example, Krupenye et al. (2017) conducted a follow-up experiment to test Heyes’s (2017) suggestion that apes who successfully completed ToM tasks in their original experiment (Krupenye et al., 2016) did so because of fundamental stimulus control mechanisms rather than through social cognitive (ToM) mechanisms. To test this hypothesis, Krupenye et al. (2017) replaced the original ape-like stimuli (humans in ape costumes) used in their original study with arbitrary, inanimate shapes (a triangle and a semi-circle) and observed the apes’ tendency to look in the location where one arbitrary shape last “saw” an object hidden. Of those apes attending to one of the two hiding places, 14 out of 22 (64%) attended to the “correct” location (where an actor, the shape in this instance, should falsely believe that the object was hidden) in the follow-up study compared to 17 out of 22 (77%) in the original study. Krupenye et al. (2017) concluded that these results suggest that stimulus-control mechanisms cannot explain apes’ ability to accurately complete such tasks. This conclusion, however, demonstrates a fundamental misunderstanding of how stimuli come to control behavior.2 The apes in this research, raised in the Wolfgang Kohler Primate Research Center, had intensive exposure to humans and other apes across their lifespans and, therefore, many examples of what humans and apes tend to do in specific contexts (including looking for things where they last saw them). The apes’ exposure to the arbitrary stimuli used in Krupenye et al. (2017), however, was limited to two familiarization trials prior to the critical tests. It is not surprising that many of the apes would not behave as if the arbitrary shape would “look for” a hidden object; these shapes differ substantially from humans and apes and, therefore, generalization should not be expected to occur to these stimuli. With no specific conditioning history, this type of study is completely inadequate for testing the stimulus control hypothesis.

Perspective Taking: Verbal Processes

Much of the behavior that we refer to as “perspective taking” is verbal in nature, but the fundamental principles outlined above do not cease to apply in such cases. If, for example, we want to understand why an observer says that a young boy who dropped his ice cream cone on the ground is “sad,” we must consider the current stimulus situation and the observer’s learning history associated with similar stimuli. Skinner (1957) provided a coherent theoretical account of how we learn to describe “private” events, those events that can only be directly experienced by one individual, such as the events that we learn to call a “stomachache” or “sadness.” According to Skinner, the individual’s community teaches them to “tact” (i.e., make a verbal response in the presence of) their own private experiences using public accompaniments to private stimuli. For example, a father who sees that his child has bumped her head and is crying (public stimuli) will often emit “tacts” (e.g., “that hurts”) related to the private events that the child is likely experiencing and reinforce “echoics” (i.e., repetition of the verbal statement, “that hurts”), which eventually leads to unprompted tacts by the child under similar circumstances. Through stimulus generalization or other processes (see below), the child may also learn to describe private events that do not have any public accompaniment. For example, she may learn to tact burning sensations from instances where she has been burned from touching something hot or caustic and then apply this tact when she feels a similar sensation internally but without any accompanying public stimuli. The child’s verbal community also reinforces their tacting of public accompaniments of private events in others. For example, when a child’s father stubs his toe, grimaces, and holds his foot, a variety of responses (e.g., “ouch,” “owie,” “hurts”) are likely to be modeled and reinforced.

An understanding of stimulus equivalence can enrich our understanding of complex verbal behavior (Guinther & Dougher, 2015), including its symbolic nature, and can shed light on the processes that lead to one’s ability to accurately tact private events to which only another person has direct access (Spradlin & Brady, 2008). Words such as “stress,” “anxiety,” and “worry” participate in stimulus classes with certain facial expressions and with behavioral patterns such as fidgeting and pacing. Observing these behaviors in others (e.g., seeing someone pacing and fidgeting) is likely to evoke one or more of the words that participate in this class of stimuli. Hearing one of the words (e.g., “I am stressing about. . . .”) is likely to lead to predictions about these and other related behavioral patterns (e.g., we might predict that they are having trouble sleeping). A child who learns to call a syringe a “syringe” may experience the stimuli associated with an injection and may also learn to tact these stimuli as “painful.” They may later learn that a “syringe” is also called a “needle.” If they are told that a lancing device for a blood test contains a needle, even if they have never had experience with the device and cannot see the needle or any blood, when seeing someone else using the lancing device, they may respond in accordance with the other person’s experience. Although the device was never directly paired with “painful” stimuli, and relevant tacts were not emitted and reinforced in the presence of those stimuli, when seeing someone else getting a blood test, the child may say that it is “painful” and also experience private, respondent behavior similar to the behavior that was elicited when they received an injection (Dougher et al., 1994; Rehfeldt & Hayes, 1998; Roche & Barnes, 1997; Valverde et al., 2009).

In addition to this type of “training” to respond to private events, children are also trained using specific narrative strategies, such as “How would you feel if your cat ran away?” Culture-consistent responses are typically prompted and reinforced. Spradlin and Brady (2008, p. 348) suggest that the ability to predict the behavior of others comes from three types of prior observations and descriptions: “1) Observation and descriptions of the behavior of a specific individual in similar situations; 2) Observation and descriptions of the behavior of many different people in similar situations; 3) Observation and descriptions of one’s own behavior in similar situations.” By taking this explicit perspective-taking training, relevant general training, and behavioral processes such as stimulus generalization and stimulus equivalence into account, we can develop and test explanations for complex verbal perspective taking without resorting to fuzzy constructs such as ToM.

Perspective Taking and RFT

Some researchers have emphasized the importance of the distinction between self and others in perspective taking and have suggested that this distinction is the basis for our ability to do perspective taking. According to Hayes (1984), "additional arbitrary contingencies” are required to perform perspective taking; the reporting of experiences must be from the locus of I or you, here or there, and now or then. For example, in the statement “I played soccer when I was young, but I play tennis now,” the "I" is constant through time even though our physical body changes. According to Hayes, this verbally3 constructed notion of the consistent locus of the speaker is a core component of perspective taking. These context-dependent words (e.g., “you,” which is only reliably effective if accompanied by other stimuli, such as the gaze of the speaker falling on a specific person) are referred to as “deictic” words in linguistics. Therefore, some researchers working within the RFT framework have dubbed this type of relational behavior “deictic framing” and have identified it as a unique “relational frame” (Barnes-Holmes et al., 2001; McHugh et al., 2004). As we will discuss shortly, this theoretical approach is problematic, but researchers have developed and tested an intervention based upon this theoretical foundation.

This perspective-taking protocol, developed by McHugh et al. (2004; sometimes referred to as the “Barnes-Holmes protocol”), consists of a list of 62 task items involving deictic components. There are three levels of complexity in the protocol: simple, reversed, and double-reversed. In the simplest trial type, the participant just needs to identify information provided in a sentence. For example, they are told, “I have a red brick, you have a green brick.” Then they are asked, “What do I have? What do you have?” In a reversed task, a conditional statement such as, “If I were you and you were me,” is provided in addition to the simple task statement, so the correct answer would be “I have a green brick” and “you have a red brick” when applied to the previous simple task example. Lastly, the double-reversed task involves another reversal as in the following example: “I am sitting here on a black chair and you are sitting there on a blue chair, if I were you and you were me and here were there and there were here, where would you be sitting?” The correct answer is “the blue chair.”

The Barnes-Holmes (BH) protocol has been used in various empirical investigations for assessing and training typically and atypically developing children (Davlin et al., 2011; Heagle & Rehfeldt, 2006; Montoya-Rodríguez & Cobos, 2016; Weil et al., 2011) and adults (Hooper et al., 2015), including those considered to have deficiencies in perspective-taking ability, such as individuals with autism (Barron et al., 2018; Belisle et al., 2016; Gilroy et al., 2015; Gómez-Becerra et al., 2007; Jackson et al., 2014; Lovett & Rehfeldt, 2014; Rehfeldt et al., 2007; Tibbetts & Rehfeldt, 2005), individuals with Down syndrome (Montoya-Rodríguez et al., 2017), social anhedonia (Vilardaga et al., 2012; Villatte, Monestès, McHugh, & i Baque, E. F.,, & Loas, G., 2010a), schizophrenia (O’Neill & Weil, 2014; Villatte, Monestès, McHugh, & i Baqué, E. F.,, & Loas, G., 2010b), and social anxiety disorder (Janssen et al., 2014).

In addition to performance on the protocol items, some researchers have measured the effects of the training on performance in other perspective-taking (i.e., ToM) assessments. So far, little evidence for the effectiveness of training with the BH protocol for improving performance in other ToM tasks exists (Jackson et al., 2014; Lovett & Rehfeldt, 2014; Montoya-Rodríguez & Cobos, 2016; O’Neill & Weil, 2014; Rendón et al., 2012; Weil et al., 2011). For instance, Jackson et al. (2014) found that none of their participants (five children diagnosed with autism) showed improved ToM scores according to the Baron-Cohen model after BH protocol training. In two experiments involving typically developing children, none of the six participants in one study (Montoya-Rodríguez & Cobos, 2016) and only one of three participants in the other (Weil et al., 2011) demonstrated improvements on other ToM tasks following training with the BH protocol. In contrast, O’Neill and Weil (2014) found that three participants diagnosed with schizophrenia who were exposed to Barnes-Holmes protocol training showed improved performance in a hinting task that was designed to assess ToM impairments. However, interpretation of results from studies evaluating the influence of the BH protocol training on ToM task performance is complicated by the finding that language skill level is positively correlated with ToM task performance in children with autism (Gómez-Becerra et al., 2007; Steele et al., 2003). When the BH protocol is effective, is this simply a function of training with complex sentence structures?

Guinther (2017) pointed out a major theoretical issue with the concept of deictic framing: deictic frames do not appear to have any of the key properties of a relational frame as defined within RFT. According to RFT, relational framing is demonstrated through mutual entailment, combinatorial entailment, and transformation of stimulus functions among stimuli participating in the relevant relational frame (Hayes et al., 2001). In line with Guinther’s observation, we have been unable to identify any mutual or combinatorial entailment associated with “deictic frames” without reference to other relational frames. As a result, it is also unclear how stimulus functions would be transformed among the stimuli participating in such a frame. By examining how these deictic “relational” stimuli (i.e., putative C_rel) have been used in the BH protocol, it is readily apparent that they are not relational stimuli but rather stimuli that can participate in other relational frames (or simply in stimulus equivalence classes). For example, “if I were you and you were me” involves an “if-then” frame and a frame of coordination. The stimuli “I” and “you” can be replaced with “Jim” and “Sally” or “the cabbage” and “the toothpick” without altering the nature of the task. The correct response, emission of one of the arbitrarily selected stimuli (e.g., the “deictic” stimuli in the original task), is determined by other relational (i.e., conditional) stimuli in the prompt. In our own (unpublished) research, we have found that similar results are obtained from a BH protocol task with deictic words and from the same task with nondeictic words. Other researchers have also modified the BH task by replacing the deictic terms with nondeictic terms, such as “yellow” and “orange” (Heagle & Rehfeldt, 2006) or “Cinderella” and “Sponge Bob” (Davlin et al., 2011; Gilroy et al., 2015), obtaining similar results in training and generalization probes as compared with results obtained in other studies implementing the original BH protocol.

Guinther (2017, 2018) proposed an alternative RFT account of derived perspective taking, the relational triangulation framework, which does not appeal to deictic relational frames, and he produced evidence in support of this account with a novel experimental task. In the first study, a demonstration of derivation of others’ “true beliefs,” seven of eight participants displayed derived coordination of the perspectives of self and others following training to establish contextually controlled “pointing” responses in relation to an observer. In the second study, analogous to the Sally–Anne task, following initial exposure to a scene in which a figurine could “see” a target item in a specific location, the target was then switched while the figurine’s “view” of the target was obstructed. Correct derivation of the “false belief” of the figurine involved responding in accordance with the target’s original location, from the figurine’s perspective. Some participants required additional training to derive “false beliefs” under testing conditions whereas others responded accurately without any additional training.

In a recent review of the BH protocol undertaken by proponents of the BH protocol, Guinther’s research was acknowledged, but his criticisms of the BH protocol were not addressed or even acknowledged (Kavanagh et al., 2020). The self-other distinction is not trivial, but distinguishing between self and others is only one of many prerequisite skills involved in perspective taking. In addition, the “deictic framing” approach to understanding this prerequisite behavior appears to be unsound. In our view, further research on the BH protocol is unlikely to significantly improve our understanding of perspective taking. If RFT has something to offer with respect to our understanding of perspective taking, we believe that this will be proportional to its contribution to our understanding of problem solving and the functions of stimuli associated with perspective-taking scenarios.

Problem Solving

A problem can be defined as a situation in which: (1) a discriminative stimulus associated with the availability of a specific reinforcer is present but (2) at least one mediating response is required before the response that produces the reinforcer can be emitted; some also add: (3) the specific sequence of responses that are required to produce the terminal reinforcer have not been reinforced in the presence of the discriminative stimulus (i.e., simple behavioral chaining is excluded; see Holth, 2008; Skinner, 1963, 1966, 1984; Epstein, 1987, 1991, 2008). Using this description, or variants thereof, we can distinguish between examples of problem-solving behavior and nonexamples, which consist of behavior that can be emitted and reinforced with the terminal reinforcer directly in response to the current stimulus conditions. For example, when you cannot find your car key or cannot open the lid of a pickle jar, you have a problem because the behavior that is required to produce the reinforcing outcome, “key in hand” or “jar opened,” cannot be produced. These would not be problems if the car key was hanging on its hook or the pickle jar could be opened by simply twisting it as usual. The combination of the discriminative stimulus for the terminal reinforcer and the “problem” stimuli serves as a compound or conditional discriminative stimulus that evokes the mediating behavior. For example, you are dressed to go to work, the clock reads “8:10,” and the key hook is empty; this stimulus combination reliably evokes a search for the car key. Depending on the specific features of the problem-solving context, different classes of mediating behavior are likely to be emitted (Holth, 2008). For example, recalling where you recently placed the pickle jar is not likely to bring you closer to opening it, but recalling where you recently placed your car key is likely to lead to locating the key. The general class of problems involving a missing item evoke a related class of behaviors that have been reinforced by finding missing items. Because problem-solving behavior is selected according to its function, it can take many forms (Skinner, 1966, 1984). Skinner (1966) described “precurrent behavior” as behavior that “furthers the reinforcement of subsequent behavior” (p. 584) and described problem solving as precurrent behavior that produces discriminative stimuli that facilitate emission of the solution.

For problem solving behavior to be evoked, the “solution” to the problem must function as a reinforcer. In the missing car key example, “car keys in hand” functions as a reinforcer because this stimulus change is required before the behavior in the larger behavioral chain of “going to work” can be emitted. Therefore, the extent to which this stimulus change functions as a reinforcer will depend upon the reinforcing effectiveness of arriving at work, or perhaps more accurately, the punishing effectiveness of arriving at work late or missing work, which can be avoided by timely departure. Motivating operations (MOs) are environmental events that alter the reinforcing or punishing effectiveness of other events and the control of behavior by stimuli associated with the response-dependent availability of those events (Edwards et al., 2019a, 2019b). MOs change the individual such that when discriminative stimuli associated with the relevant reinforcer are encountered, these stimuli evoke behavior that was previously reinforced in their presence. MOs, therefore, are another factor that must be considered when analyzing problem-solving behavior. When the required precurrent behavior does not occur, or when it occurs only briefly and then the individual “gives up,” the solution to the problem may not be adequately reinforcing at that moment.

The behavioral processes associated with chaining are relevant to many problem-solving scenarios. In Epstein’s (1981) well-known experiment, pigeons “spontaneously” solved the problem of reaching a suspended banana by emitting components of the required sequence of responses that had been trained separately: moving a box to a specific location and standing on the box. In test trials, the stimulus context consisted of a banana suspended out of reach and a box positioned too far from the banana. Although, in training, the pigeons moved the box toward a target on the wall, the problem context with no such target was sufficiently similar to the training context to evoke the behavior required to move the box to the appropriate location, beneath the banana. The box positioned beneath the banana had been encountered during training already, so it evoked mounting the box, which enabled banana pecking. Therefore, as in a behavioral chaining example, each intermediate stimulus change reinforced the behavior producing it. Unlike a standard behavioral chaining example, however, the specific problem context had never been presented, and the necessary sequence of behaviors was never previously reinforced. The pigeon’s history of conditioning resulted in generalization of successful precurrent behavior in the novel problem context. As described in Part 3 of our definition of a problem, novelty may be considered as a requirement for a context to be classified as a “problem.” The problem context may serve as a testing ground for stimulus functions that have not been directly conditioned; these stimulus functions may arise through stimulus generalization or through other processes, such as those underlying stimulus equivalence.

We can increase the chance of finding a remote control by changing our position in the environment (e.g., kneeling down, changing our orientation, or getting closer to a location) or by manipulating the environment (e.g., looking under or between cushions). We can learn to use symbols as S^Δ to eliminate unnecessary responses (as on a diagnostic flowchart) or as S^D to guide additional behavior (Skinner, 1984). We can use tick marks to indicate that an object has been inspected, use tally marks when counting, or alphabetize a list of names when searching for specific names. We can also generate discriminative stimuli by thinking; these generated stimuli may evoke additional precurrent behavior or the solution to the problem itself. Thinking is operant behavior and can be analyzed as such (Moore, 2015), but access to the relevant behavior presents some additional challenges, in particular when it is private (see Hayes, 1986; Moore, 2000; Rachlin, 2018). When solving problems, private verbal behavior (a common form of thinking) can serve as precurrent behavior that facilitates production of the solution to the problem. Reproduction or generation of rules, an example of thinking, can be particularly useful in the appropriate context. For example, while disassembling an apparatus, repetition of the rule, “righty-tighty, lefty-loosey,” may facilitate the removal of a component. When solving a puzzle, generation of rules such as, “rotating this part moves that part,” can facilitate production of the solution.

Imagining is perceptual behavior in the absence of the thing perceived (“seeing in the absence of the thing seen,” Skinner, 1953), and such behavior is a common form of precurrent behavior. Imagining is necessarily covert (without appropriate instrumentation) because, unlike thinking involving verbal behavior, there is no public correlate of such behavior; it is what we do when listening, seeing, and otherwise sensing the environment, whether or not the relevant features of the environment are present or absent. With exposure to a sufficient number of exemplars, we may learn to imagine rotated objects or features of a landscape from different vantage points with great accuracy. These abilities are often shaped in practical problem-solving scenarios, such as those involved with operation of robotic equipment in medical teleoperation, underwater exploration, and mechanical work in outer space (Menchaca-Brandan et al., 2007), or those involved with playing computer games (Granic et al., 2014), but they may also be shaped by such “mundane” activities as navigating our everyday environments and manipulating commonly encountered objects.

Recall is a category of behavior that is particularly relevant to problem solving. Accurate recall entails responding in a way that corresponds with a past event when features of that event are absent in the present context. For example, we can answer questions about what we ate for dinner last night or what we did last weekend, even though the dinner contents and the weekend activities are not directly in front of us. From a behavioral perspective, questions about the past or other situations under which recall is likely to be reinforced are regarded as discriminative stimuli for such behavior (Baum, 2011; Catania, 2013; Epstein, 2014; Moore, 2015; Palmer, 1991; Skinner, 1963). These discriminative stimuli in combination with the temporally distant stimuli that correspond with the events to be recalled serve as the ultimate sources of control for recalling but, as just described, recalling is often controlled by stimuli that are generated by additional behavior, such as thinking and imagining, that is emitted in the recall process. When responding to a question, recall is typically reinforced with social consequences, but when solving a problem, successful recall is reinforced, at least sometimes, with the solution to the problem. Successfully answering a simple question often qualifies as problem solving because, in order to emit the correct response, such as “gardening” in response to “What did you do this weekend?,” some additional precurrent behavior may be required (e.g., saying to yourself, “On Tuesday, I went to yoga after work, then when I got home. . .”). An example of recalling that does not qualify as problem solving is immediately responding “nine” when asked “What is three times three?” In this case, the response is under the direct stimulus control of the question.

Stimulus equivalence and related phenomena are directly relevant to problem solving. Stimuli that participate in an equivalence class, including those stimuli that share no formal similarity with other stimuli in the class, can have adaptive stimulus functions that are not directly conditioned. In a problem-solving scenario with little physical similarity to previously encountered scenarios, this additional source of stimulus control (beyond the stimulus control arising from stimulus generalization, which is dependent upon formal similarity) may significantly enhance the problem solver’s chances of success. For example, when navigating to a restaurant in a large hotel, the brand (i.e., symbol) for the restaurant might appear on an elevator button. As a result of seeing the symbol beside a description of the restaurant on a sign in the lobby earlier, the symbol may evoke pressing of the correct button, leading to access to the restaurant, even though pressing buttons with that symbol has never previously gained the individual access to a restaurant. As described previously, stimulus equivalence appears to play an important role in verbal behavior. The relationship between stimulus equivalence and verbal behavior appears to be bidirectional, with one facilitating the other, but the behavioral mechanisms underlying stimulus equivalence are not completely understood. There are some promising theories on this topic (Greer & Longano, 2010; Hall & Chase, 1991; Rehfeldt & Hayes, 1998).

RFT is a theoretical account that incorporates both equivalence and nonequivalence relations such as “distinction” and other logical operations that are clearly relevant to problem solving (Hayes et al., 2001). RFT specifies that relational responding (including stimulus equivalence, one example of relational responding) is a generalized operant that results from a history of multiple exemplar training, but this explanation of relational behavior has been criticized as oversimplified (Gross & Fox, 2009). “Accurate” relational responses are often produced after a sequence of precurrent behavior. Therefore, trials in relevant experiments or circumstances in which relational responding might occur, may meet the definition of a “problem,” with the relational response being the outcome of successful problem solving (Diaz et al., 2020; Miguel, 2018; Moustakis & Mellon, 2018; Palmer, 2004). Explaining problem solving with another specific type of problem solving (i.e., relational responding) may represent a valid higher-order explanation, but the extent to which this type of analysis is useful, in particular in the absence of an explanation for the problem solving that occurs in the process of relational responding, is unclear to us. Putting these issues aside, we will briefly summarize the potential contribution of RFT to our treatment of problem solving, according to our understanding of RFT as it relates to problem solving.

Hayes et al. (2001) described verbal problem solving as “pragmatic verbal analysis that changes the behavioral functions of the environment under the antecedent and consequential control of an apparent absence of effective action” (p. 96). The verbally abstracted “goal” and other stimuli that are relevant to a problem-solving situation participate in a wide variety of relational frames. For example, the verbally abstracted goal may participate directly or indirectly (e.g., via frames of coordination) in several “if–then” frames. By engaging in relevant relational behavior, the behavioral functions of a variety of stimuli may be transformed. For example, when solving the problem of getting a mobile phone signal in the wilderness, the relational networks associated with “if we gain elevation, we may get a signal,” change the functions of stimuli that are related to elevation gain; the sight of a hill may serve as a reinforcer and may evoke climbing because of the relationship between frames of comparison involving elevation and frames of comparison involving signal strength. These comparative relations may result in higher hills having stronger reinforcing and evocative functions. This higher-level approach to understanding problem solving and other complex behavior is appealing because it appears to account for complex problem-solving behavior, including behavior that is otherwise difficult to explain (i.e., completely unorthodox solutions). RFT was developed, in part, to explain rule-governed behavior and to extend our understanding of verbal behavior, so these areas of inquiry are highly interrelated, as is apparent in the previous example (McLoughlin & Stewart, 2017; McLoughlin et al., 2020).

There are many other important pieces of the problem-solving puzzle, but this brief overview is sufficient for the purpose of supporting our next point. We now return to perspective taking and make the case that our efforts to understand perspective-taking behavior should be combined with our efforts to understand problem-solving behavior because these two categories of behavior have considerable overlap.

Perspective Taking as Problem Solving

An examination of the perspective-taking literature reveals that researchers have primarily investigated perspective taking using tasks that require problem solving. For example, in false belief tasks, such as the “Sally–Anne” task, the participant is exposed to a sequence of stimuli and finally prompted to respond (e.g., “Where does Sally think the marble is?”). Accurate responding depends on the participant’s ability to recall where the marble was when Sally was present, in this example. Thus, some precurrent behavior is required to “solve the problem.” In VPT tasks, such as the three-mountain task, the participant’s selection of the picture that corresponds with a specific vantage point is reliant upon attending to features of the landscape and features in the pictures, and it may involve a series of covert responses (e.g., tacting features in the pictures that do or do not match the relevant vantage point). Likewise, with the unexpected identity task, the participant must recall or look to see what is advertised on the container when responding to questions about what others will think is in the container.

Successful completion of the tasks in the BH protocol also appears to rely upon precurrent behavior. Initial accurate performance in this task is dependent upon significant precurrent behavior in response to the key words associated with each trial type, in particular in reversed and double-reversed trials. With additional exposure to the protocol, “accurate” responses corresponding with the simple, reversed, and double-reversed conditions may come to be controlled by other stimuli, including the length of the prompt in the trial, because this is reliably correlated with the trial type (double-reversed trials have the longest prompt). With significant training, to solve the tasks, the participants need only attend to these aspects of the prompt and the arbitrary stimuli that participate in the initial grammatical frame (e.g., I have a red block, you have a blue block) and they can respond appropriately (e.g., with the longest prompt—a double-reversal—responding in accordance with the original, unchanged, statement will always be correct). In our own (unpublished) research with this protocol, participants reported using these “patterns” to respond so that they could save time answering “repetitive” and “similar” questions. In this way, the problem becomes simpler with additional exposure to the task. Even so, it appears that precurrent behavior, such as looking back at or recalling the initial statement, is still required to respond accurately.

Problem solving in perspective-taking tasks often relies upon control by stimuli that reliably predict the behavior of others. For example, in the “Sally–Anne” task, the predictive stimuli are the presence of Sally and the placement of the marble in one location; Sally looking in that same location is the behavior that reliably occurs under those stimulus conditions. Therefore, in many cases, perspective taking is solving problems involving prediction of what others will do (or are currently doing privately). “Do” should be construed broadly to include what others will think or say (or what they are thinking or imagining privately), where they might go next, how they will react to questions or other stimuli (like the sight of a spider), and so on (see Guinther, 2017, 2018). Problem solving in VPT tasks may involve control by stimuli that are predictive of others’ or one’s own perceptual experience if they were in a different location.

In perspective-taking tasks, “others” can include other organisms, dolls, or other inanimate objects, and hypothetical alternative positions of the self and other organisms or objects. The “solution” to perspective-taking problems (e.g., the “correct” answer in the Sally–Anne task) requires that the individual behave in accordance with the “other” having at least perceptual, but often greater, behavioral capabilities. Nevertheless, describing perspective taking as behavior that accords with others having perceptual or other behavioral capabilities does not necessitate that the perspective taker is engaging in, or capable of engaging in, verbal behavior to this effect. This description of perspective taking is reflective of the conditions under which others are likely to say someone is demonstrating perspective taking, but it says nothing of the behavioral processes that are involved in the production of such behavior. If an ape accurately predicts where a human in an ape costume will look for food, we can call this perspective taking without concluding that the ape is describing to itself, or generating an internal representation of, the perspective of the human.

In perspective-taking “problems,” there are behavioral prerequisites that must be met for a participant to respond accurately. With the Sally–Anne task, these include an appropriate verbal repertoire, the ability to recall details of a story, and a history of reinforcement for predicting that people look for things where they last saw them. As with Epstein’s pigeons or a mathematician solving a complex problem, the perspective-taking context evokes relevant precurrent behaviors; these behaviors in turn produce stimuli that enable the individual to solve the perspective-taking problem. Examples of skills that are common prerequisites for solving perspective-taking problems include accurate recall; tacting of facial expressions and other public correlates of private events (e.g., public correlates of “pain”); social referencing, including appropriate responding to gestures and eye gaze; and an adequate speaker and listener repertoire.

It is important to note that, as when studying problem solving, researchers administering perspective-taking tasks work under the assumption that participants have not been exposed to the specific experimental task. Otherwise, it would not be possible to draw any meaningful conclusions about a participant’s ability to solve such problems (i.e., do perspective taking) in general. If a child were repeatedly exposed to the Sally–Anne task, and the response, “the box,” was reinforced whereas the response, “the basket,” was not, the correct response could be emitted in direct response to the final question without the participant engaging in any precurrent behavior (if “the box” was always the correct answer). Following such training, if the task were changed slightly such that the other location was correct, the participant would respond incorrectly. Therefore, for perspective-taking problems, such as those typically used to study perspective-taking behavior, emission of relevant precurrent behavior in a novel context is also a critical prerequisite for successful perspective taking. Emission of adaptive precurrent behavior in a novel context may occur as a result of stimulus generalization. But as described in our discussion of problem solving, transfer (or transformation) of the functions of stimuli that participate in equivalence (or other) stimulus classes can result in the control of precurrent behavior by stimuli in the novel context, even when those stimuli have no formal similarity to discriminative stimuli for useful precurrent behavior in previous circumstances. As a result of these processes and the interrelated processes associated with rule-governed and verbal behavior in general, a novel perspective-taking context may be a treasure trove of stimuli that evoke sequences of productive precurrent behavior, leading to solutions to the problem.

By treating perspective taking as problem solving, we must also reconsider the self-other distinction, which has been treated as a core element of perspective taking by researchers working under the prominent deictic relational frame approach within the RFT framework. The self-other distinction is not relevant in many examples of perspective taking and is just one prerequisite skill that is required to solve other perspective-taking problems. For example, in the guesser–knower task, the participant does not know the location of the object, so they do not need to discriminate between their own and others’ knowledge. Instead, they must discriminate between the guesser and the knower based on the stimuli associated with their gaze (and, of course, they must discriminate between them in general when solving the task). In some perspective-taking tasks, such as the Sally–Anne task, the participant must discriminate between stimuli to which they and others have access, but this is just one of many prerequisites for success with this task. The behavioral principles that are associated with discriminating between stimuli to which the self and others have access (i.e., events that they have experienced or are experiencing) do not differ from the principles associated with discriminating between the experiences of two different “others.” The stimuli associated with the self-perspective are qualitatively but not functionally different from those associated with the others-perspective (Moore, 2015). For example, Lattal (1975) and Shimp (1983) trained pigeons to report separately “what they did” and “what they said they did,” effectively demonstrating self-awareness capability in pigeons, which suggests that fundamental stimulus control mechanisms can parsimoniously explain self-awareness. Nevertheless, “concept of self” or “self-awareness” is an important phenomenon with relevance to perspective taking and many other aspects of behavior.

Although we have focused on examples of experimental tasks designed to study perspective taking in this analysis, much of everyday perspective taking also appears to be accurately described as problem solving. When playing chess or go,4 definitive examples of perspective taking, aside from well-practiced opening moves, much of the game play is filled with pauses during which the players are covertly exploring the likely outcomes of many different moves, including their opponent’s likely responses. When negotiating with a partner or a potential client, thinking about what the other person wants and what they value is often an important prerequisite for presenting a satisfactory suggestion. Choosing a “good” birthday gift for a friend involves thinking about their interests and recalling earlier conversations. Precurrent behavior appears to play an important role in these and most other examples of perspective taking. However, problem scenarios that do not require much or any precurrent behavior, such as the identification of emotional states (in accordance with the standards of the verbal community) may represent examples of perspective taking that should not be classified as problem solving. These “simple” perspective taking skills may be conceptualized as prerequisites for solving more complex perspective-taking problems.

Applying the Problem-Solving Approach

Perspective-taking deficits may be the result of broader problem-solving deficits or they may be specifically associated with inadequate control of behavior by relevant stimuli, or both. Determining where the underlying deficit lies is critical to developing an appropriate intervention. Solving perspective-taking problems is usually dependent upon control of behavior by antecedent social stimuli, including subtle facial expressions and other stimuli produced by the behavior of others, often in combination with other nonsocial elements of the environment. The functions of social stimuli, therefore, should be considered when troubleshooting perspective-taking deficits. Inadequate control by social stimuli has been implicated in perspective-taking deficits in various populations, in particular people diagnosed with ASD. For example, Gale et al. (2019) found that children who were diagnosed with ASD preferred nonsocial stimuli (videos of abstract geometric patterns) over social stimuli (videos of children’s and dogs’ faces), whereas developmental-age-matched controls showed no difference in preference between the two classes of stimuli. In addition, the outcomes that serve as reinforcers for successful perspective taking are frequently social stimuli, such as approval from others. Therefore, limited capacity for social stimuli to serve as discriminative stimuli and reinforcers represents a significant barrier to successful perspective taking. Although there are some promising approaches to assessing and addressing social stimulus deficits, we still have much to learn (Rodriguez & Gutierrez, 2017).

Success in many problem-solving tasks, including perspective-taking tasks, relies upon simultaneous control of behavior by multiple features of the environment. Stimulus overselectivity, therefore, represents a general problem-solving deficit. Stimulus overselectivity is also commonly observed in individuals diagnosed with ASD (Ploog, 2010). Verbal behavior deficits also represent a general deficit. Many general problems and perspective-taking problems are specifically verbal in nature, but problems that are not specifically verbal in nature may be solved more readily by verbally capable participants because of their ability to engage in a wide variety of precurrent behavior, including production of stimuli that participate in large networks of stimulus classes (and even asking for help). It is clear that verbal behavior deficits would place an individual at a disadvantage when solving problems, including those of the perspective-taking variety. There are certainly other general problem-solving deficits that would limit an individual’s ability to engage in perspective taking, but we will not attempt to provide an exhaustive account here. One fruitful approach to determining if a specific perspective-taking deficit stems from social stimulus issues or from more general problem-solving deficits may be to develop a task that is analogous to the perspective-taking task in question but does not involve social stimuli. If, given an appropriate training history, the individual can solve the task, this would demonstrate that the relevant precurrent behavior is available but is not being evoked (or adequately reinforced) by conventional social stimuli. In such cases, through generalization training or establishment of relevant stimulus classes, the precurrent behavior that is necessary to solve the perspective-taking problem might be brought under conventional stimulus control to address the perspective-taking deficit.

Of course, application of our technology is not limited to addressing deficits; we may also apply our understanding of perspective taking to improve early education or to enhance typically developed adults’ ability to engage in complex perspective taking. These applications are also socially significant and, therefore, worthy of consideration. By treating perspective taking as problem solving with social stimuli, we are directed to consider and evaluate the functions of relevant stimuli for the individual; thus (1) consider the specific precurrent behaviors that are required to successfully complete a given perspective taking task, similar to conducting a task analysis; (2) confirm that the individual is capable of emitting the specific precurrent behaviors; (3) bring the precurrent behaviors under relevant stimulus control; (4) test for generalization to novel perspective-taking contexts; and (5) as necessary, program for generalization and/or establish stimulus classes that lead to control of requisite precurrent behavior in a variety of relevant perspective-taking contexts. This problem-solving approach to perspective taking does not invalidate promising behavioral research that has already been undertaken in this area, for example, work on establishing perspective-taking prerequisites (e.g., Gould et al., 2011; LeBlanc et al., 2003). Instead, this approach may help to clarify the specific contribution of existing research to a broader understanding of perspective taking.

Conclusion

Our goal with this analysis is not to identify every aspect of problem solving that is relevant to perspective taking but instead to call attention to the overlap between these two general classes of behavior and to highlight what appears to be a clear advantage of treating perspective taking as problem solving in many cases. Treating perspective taking as problem solving does not “solve the problem” of perspective taking. We still have much to learn about problem solving, and we propose that advances in this area would lead directly to improvements in our understanding of perspective taking. For example, as we discussed earlier, our understanding of the role of stimulus equivalence and relational framing in problem solving is still limited. But a general framework for analyzing problem solving and a good experimental foundation are already in existence. We suggest that this framework and this foundation are an appropriate starting point for a scientific understanding of perspective taking.

Data Availability

Not applicable

Declarations

Conflicts of Interests

We have no conflicts of interest to disclose.

Code Availability

Not applicable

Ethics Approval

Not applicable

Consent to Participate

Not applicable

Consent for Publication

Not applicable