What do we know about implicit false-belief tracking?

There is now considerable evidence that neurotypical individuals track the internal cognitions of others, even in the absence of instructions to do so. This finding has prompted the suggestion that humans possess an implicit mental state tracking system (implicit Theory of Mind, ToM) that exists alongside a system that allows the deliberate and explicit analysis of the mental states of others (explicit ToM). Here we evaluate the evidence for this hypothesis and assess the extent to which implicit and explicit ToM operations are distinct. We review evidence showing that adults can indeed engage in ToM processing even without being conscious of doing so. However, at the same time, there is evidence that explicit and implicit ToM operations share some functional features, including drawing on executive resources. Based on the available evidence, we propose that implicit and explicit ToM operations overlap and should only be considered partially distinct.

Theory of Mind (ToM) reasoning or mentalizing refers to an individual's ability to infer the mental states of others, including their beliefs, feelings, and intentions (Apperly & Butterfill, 2009;Butterfill & Apperly, 2013;Frith & Frith, 2005;Low & Perner, 2012). ToM is a complex and dynamic cognitive process that is engaged across a wide range of social activities. Cooperating and communicating with work colleagues, interacting with family and friends, thinking about others in their absence, or simply asking a stranger for help all require ToM. Underscoring the importance of ToM for social functioning are the social-communicative limitations seen in individuals with an autism spectrum disorder (ASD) or schizophrenia, who typically show impairments, relative to neurotypicals, in ToM reasoning (Baron-Cohen, Leslie, & Frith, 1985;Brüne, 2005;Frith, 2004a, b;Frith & Hill, 2004;Moran et al., 2011). Further, across the neurotypical human lifespan, ToM abilities have been shown to predict social functioning outcomes (Apperly, Samson, & Humphreys, 2009;Dumontheil, Apperly, & Blakemore, 2010;Henry, Phillips, Ruffman, & Bailey, 2013;Maylor, Moulson, Muncer, & Taylor, 2002;Phillips et al., 2011;Slaughter, Peterson, & Moore, 2013). Currently, there is considerable debate regarding the cognitive architecture underlying ToM. A recent, and particularly provocative claim, is that neurotypical individuals can not only reason about others' beliefs, but also implicitly track them as social events unfold. That is, humans have the ability to register what is represented in another person's mind even in the absence of an intention to do so and without explicit knowledge of doing so. This claim entails a distinction between implicit and explicit ToM functions.
Much of the research that has prompted the proposal of implicit ToM comes from the field of developmental psychology. Specifically, by assessing indirect measures such as eye movements, researchers have concluded that infants as young as 7-15 months are able to register false beliefs (i.e., they can recognize that others can have beliefs about the world that are different from reality, based on outdated knowledge; Kovács, Téglás, & Endress, 2010;Onishi & Baillargeon, 2005;Surian, Caldi, & Sperber, 2007). These findings have prompted a revision of thinking about the developmental trajectory of ToM, since more than 20 years of research employing explicit measures of ToM (e.g., verbal responses, pointing) indicated that accurate attribution of others' false beliefs was not reliably demonstrated until children reach approximately 4 years of age (Wellman, Cross, & Watson, 2001). Thus, along with the proposal for an implicit-explicit ToM dimension, the results from the infancy studies suggest the existence of early and later developing ToM capacities, which may or may not be developmentally continuous (Baillargeon, Scott, & He, 2010;Perner & Roessler, 2012). In addition, such findings have led to the hypothesis that along the dichotomy of implicit/explicit ToM functions, two opposing ToM systems coexist in adults: that is, a rapidly operating, resourceefficient, and inflexible system, as well as a slower operating, resource-demanding, and flexible ToM system (Apperly & Butterfill, 2009;Back & Apperly, 2010). This review will outline and assess current evidence for and against the hypothesis of an implicit-explicit ToM distinction. To organize our review, we introduce the origins of the concept of implicit ToM reasoning, summarize terminology that has been used to describe this new form of ToM, and evaluate assumptions made regarding its underlying mechanisms. In doing this, we also explore relationships between the proposed implicit and explicit ToM systems and examine the extent to which they draw on overlapping cognitive resources.

Origin of the concept "implicit" Theory of Mind
In their now classic work, Clements and Perner (1994) demonstrated that children by the age of 2 years and 11 months show sensitivity to the beliefs of others in a standard false-belief task (sometimes also referred to as the "Sally-Anne" paradigm or the location-change false-belief test, originally devised by Wimmer & Perner, 1983). In this paradigm, a protagonist (e.g., Sally) places an object (e.g., a ball) in one location (e.g., a basket) and then leaves the scene (e.g., she goes out of the room). In the protagonist's absence, another individual (e.g., Anne) moves the object to a different location (e.g., a box) and then leaves the scene. (See Fig. 1.) Participants are then asked to report where the protagonist will look for the object when she returns.
To successfully pass this test, participants must understand that the protagonist's actions will be based on what she believes to be true, rather than the actual state of affairs. Thus, participants should indicate that Sally will look for the object in the basket rather than the box. Typically, participants' responses to the Sally-Anne test question are measured using direct tests (e.g., verbal responses or pointing). In their groundbreaking variation of this task, Clements and Perner (1994) recorded children's anticipatory gaze behavior toward the two locations in the scene as well as asking the explicit test question. For the anticipatory gaze behavior, one would expect participants to fixate first on and look longer at locations that are consistent with the beliefs and anticipated behavior of the protagonist (e.g., Sally moving toward the location that is in line with her false belief about the ball, thus the empty basket). Clements and Perner found that 90 % of child participants looked at the location that was consistent with the protagonist's false-belief belief state; however, only 55 % of them could explicitly indicate this location in answering the explicit test question. Clements and Perner interpreted the differences between children's performance on the indirect looking measure and their answers to the direct question as evidence for distinct implicit Fig. 1 "Sally-Anne" false-belief test: A protagonist (e.g., Sally, left character) places an object (e.g., a marble) in one location (e.g., a basket), while another individual (e.g., Anne, right character) watches. Sally then leaves the scene (e.g., she goes out of the room). In Sally's absence, Anne now moves the object to a different location (e.g., a box). She then also leaves the scene and Sally returns. At this point in time participants are asked to report where Sally will look for the object. If participants understand that her actions will be based on what she believes to be true, rather than the actual state of affairs, they should indicate that Sally will look for the marble in the basket rather than the box. Sally will act on her false belief. Artist: Axel Scheffler, reused with permission from Elsevier. From Frith, U. (2001). Mind blindness and the brain in autism. Neuron,32,[969][970][971][972][973][974][975][976][977][978][979][980] and explicit ToM systems ("…the signs are that we are dealing with a different type of knowledge; implicit as opposed to explicit knowledge…", p. 394, Clements & Perner, 1994). Clements and Perner (1994) were the first to empirically investigate a difference between implicit and explicit ToM. Their investigations intended to tap the "consciousness" aspects of ToM processing (i.e., the divergent moment for the ability to actively reflect and report on ToM processes and behavior; Low & Perner, 2012), building on work from the field of memory (for a review, see Reber, 2013). Explicit memory relates to the operation of intentionally and consciously remembering information, such as particular facts. In contrast, implicit memory is the ability to acquire information without conscious knowledge of doing so, such as how to perform a specific task. Classic examples of implicit memory are tying a shoe or riding a bicycle (i.e., these types of tasks can be undertaken by people without remembering facts and details about the action procedure). In line with this distinction, implicit ToM is the process of representing another's mental state without conscious access to this information.
The key findings of Clements and Perner (1994) have now been replicated by several groups, further supporting the hypothesis that looking measures in false-belief tasks can tap unconscious ToM processes (Garnham & Perner, 2001;Low, 2010;Ruffman, Garnham, Import, & Connolly, 2001). In a compelling study by Ruffman et al. (2001) using a Sally-Anne task, 3-to 5-year-old children had to bet on where a story character would look: either the location consistent with the protagonist's false belief or the actual location of the object. The investigators reasoned that betting is a type of confidence measure, which, like a verbal or pointing behavior, requires explicit ToM processing. However, unlike pointing or verbal responses to a test question, both of which demand a dichotomous response, betting allows an indication of greater or lesser certainty regarding an answer. As such, the Ruffman et al. study investigated whether looking behavior actually indexes explicit knowledge of another's belief state, but held with little confidence, by studying looking and betting behavior at the same time. If the looking measure actually reflects an explicit ToM process, children's betting should be consistent, reflecting the wide range of measurements for explicit falsebelief understanding. By contrast, if looking behavior indexes a distinct system of earlier-developing, implicit knowledge, then children's betting should be unrelated to their eye-gaze patterns. The authors found that children consistently bet on the actual location of the object rather than the location represented by the false belief of the character. At the same time, they reliably looked to the false-belief location. This was taken as further evidence that looking measures are distinct from explicit false-belief reasoning, and therefore represent an implicit/unconscious ToM process.
Recent work with adults further supports the proposal that humans have the capacity to represent unconsciously the cognitions of others. Adapting a paradigm developed by Senju, Southgate, White, and Frith (2009), Schneider, Bayliss, Dux, and their colleagues used a false-belief anticipatory looking paradigm, as described above, utilizing eyetracking technology. In addition, the authors also included an offline, follow-up debriefing to assess the extent to which adults were conscious of having engaged in ToM processing (Schneider, Bayliss, Becker, & Dux, 2012;Schneider, Lam, Bayliss, & Dux, 2012;Schneider, Slaughter, Bayliss, & Dux, 2013). Specifically, Schneider and colleagues used movies to display a Sally-Anne type task, in which the protagonist in one scenario held a false belief and in the other a true belief about the location of an object. True-belief scenarios served as control conditions and differed from false-belief scenarios in that the protagonist knew at the end of the trial sequence where the object was located. In the task used by Schneider and colleagues, at the end of each video sequence, for both the false-and true-belief scenarios (i.e., once the protagonist was back in the room), the last shot of the sequence was displayed for another 5 sec to allow measurement of participants' anticipatory looking behavior. Crucially, in this time window, participants looked at the empty location (i.e., the location that contained no ball) for longer in false-belief than in truebelief conditions. Note that in the false-belief condition, the protagonist falsely believed the ball to be at the empty location, whereas in the true-belief condition she correctly believed the ball not to be at that location. Thus, participants demonstrated processing the belief state of the protagonist. In addition, a seven-item debriefing procedure (adapted from Bargh & Chartrand, 2000) was administered after the eye-tracking sessions, probing participants with increased specificity regarding their explicit/conscious insight into having engaged in a ToM process. This approach was sensitive enough to identify participants that were conscious of having tracked the belief states of the displayed protagonists (e.g., they would report that they noticed that "Sally" had been tricked). Importantly, participants that showed no conscious registration of belief tracking nevertheless displayed eye movements consistent with engaging in belief processing. Thus, adult participants displayed eyemovement behavior indicative of false-belief processing, without being conscious of such behavior.
Another key aspect of the study by Schneider, Bayliss et al. (2012) was that they employed a multiple-trial design. In many previous studies on implicit false-belief tracking, it was common practice to test looking behavior on a single trial only (because of the fact that most of the research work stemmed from the infant/child literature). However, social interactions are dynamic and often prolonged. Therefore it was important to confirm that the eye-tracking results reflected the implicit analysis of others' mental states, rather than a social orienting response, such as that triggered, for example, by biological motion (Allison, Puce, & McCarthy, 2000). Further, it helped to demonstrate whether eye-movement evidence for false-belief tracking simply reflected behavioral rule learning, rather than a representation of mental processes (e.g., Perner & Ruffmann, 2005;Ruffman, Taumoepeau, & Perkins, 2012). The multiple-trial design used by Schneider, Bayliss et al. (2012) assessed the time course of false-belief looking behavior. The authors reasoned that if a social orienting response was responsible for the previously observed implicit ToM eye-movement findings, then fast habituation of these effects should be evident (as found for orienting processes; Asplund, Todd, Snyder, & Marois, 2010;Sokolov, Spinks, Näätänen, & Lyytinen, 2002). Similarly, a behavioral rule account would predict that the typically observed eye-movement behavior would take several trials to develop. In contrast, if a social analysis/mental representation process was being measured in implict false-belief tasks, then it was predicted that the eye-movement behavior would be observed immediately and sustained over multiple trials. In support for a social analysis/mental representation account, two experiments revealed that from the first trial, looking behavior consistent with false-belief tracking occurred and was sustained over a 1-hour period. Thus, it can be concluded with some confidence that for children and adults, false-belief tracking without instruction reflects ongoing social analysis that occurs without a conscious registration of this behavior.
In addition to anticipatory looking as a measure of falsebelief understanding, one often sees violation-of-expectation (VoE) paradigms employed (e.g., Onishi & Baillargeon, 2005;Surian et al., 2007). Here, participants' fixations are examined in relation to unexpected events in false-belief scenarios. In these VoE studies, implicit belief processing is inferred if participants show longer fixation times to scenarios in which protagonists behave in opposition to their belief (e.g., Sally moves toward the ball's actual location instead of the location where she believes it to be). For example, in line with Schneider, Bayliss et al. (2012) and the social analysis/ mental representation account, Yott and Poulin-Dubois (2012) found, using a VoE false-belief paradigm, that 18month-olds displayed eye-movement behavior consistent with having learned a new rule (i.e., an object that disappears in location A can be found in location B). Despite this, when tested with a false-belief scenario, infants kept displaying the opposite pattern of eye movements, consistent with them having represented the belief state of a displayed protagonist. In further support of the social analysis/mental representation account of eye-movement behavior in false-belief tasks, Senju, Southgate, Snape, Leonard, and Csibra (2011) had one group of 18-month-olds experience an opaque blindfold and another one experience a transparent "trick" blindfold. Both groups then observed an actor wearing the blindfold they had just experienced while a puppet moved an object away from its location. This meant that only in the opaque blindfold group was a false-belief condition established, because in the transparent blindfold condition, infants should have inferred that the protagonist was always able to see. Interestingly, this was reflected in the infants' anticipatory eye-movement behavior. If eye-movement behavior could be explained by situation or task factors, such as behavioral rules or statistical learning, one would have expected the same anticipatory eyemovement behavior in both conditions. Work with individuals with high cognitive abilities and a less severe autism spectrum disorder (ASD; Senju et al., 2009) has provided additional evidence for the proposal of distinct implicit and explicit ToM systems. Specifically, Senju et al. (2009) found that adults with ASD passed explicit false-belief tasks, such as the Smarties task (i.e., in a box of Smarties, a participant finds pencils and is then asked what another person, who has not yet seen the contents of the box, would indicate to be in the box; Perner, Leekam, & Wimmer, 1987) or the Strange Stories task (i.e., naturalistic social stories, which concern the different motivations that can lie behind everyday events/utterances that are not literally true (e.g., being polite to spare somebody's feelings); Happé, 1994), as previously found in other studies (Bowler, 1992;Peterson, Slaughter, & Paynter, 2007;Scheeren, de Rosnay, Koot, & Begeer, 2013). However, when probed on a false-belief anticipatory looking paradigm, these same participants failed to display eye-movement patterns consistent with implicit ToM processing. This pattern of results was later confirmed and extended by Schneider et al. (2013), who demonstrated that even when tested over a prolonged time period, individuals with ASD did not demonstrate belief processing in anticipatory false-belief tasks, while passing explicit ToM measures. Thus, evidence is mounting that implicit and explicit ToM processes are indeed distinct.
With the intention of further teasing apart what distinguishing characteristics underlie implicit and explicit ToM, Low and Watts (2013) recently reported an elegant adaptation of the Sally-Anne paradigm. The authors used a classic false-belief location-change task and a novel falsebelief identity-change task in both an implicit and explicit task setup. The implicit and explicit task setup was realized with anticipatory looking and a verbal/pointing response, respectively, toward the false-belief location. The false-belief identity-change task presented a protagonist and an object that had two apparent identities (i.e., a red and a blue side to a toy robot). Here, the red side of the toy robot was shown to the participants as it moved from one box to another (left to right from the participant's perspective). Once the toy robot reached the right box, it turned around (visible only to the participant), which revealed that it had two sides: a red one and a blue one. The toy robot then returned to the initial box, now showing its blue side to the participant. At this time point the participant would have recognized that the protagonist would have seen with the first movement (to the right box) the blue side of the toy robot and with the second move (to the left box) the red side of the toy robot. Thus, participants should infer that the protagonist believes that there are two objects in the scene (i.e., red and blue toy robots). In regard to anticipation, participants learned in previous familiarization trials that the protagonist had a preference for blue objects. Thus, in the implicit and explicit test, participants should have anticipated/ indicated that the protagonist would reach toward the location where he/she falsely believed the blue toy robot to be (i.e., the right box). The authors found that in the false-belief locationchange task, adults, 4-year-olds, and the majority of 3-yearolds showed anticipatory looking in line with the false belief of the protagonist. Further, they found that accuracy of explicit responses increased as a function of age (3-year-olds: 31 %; 4year-olds: 75 %; adults: 100 %). On the implicit and explicit false-belief identity-change task, however, the majority of participants in all age groups did not show anticipatory looking in line with the false belief of the protagonist (3and 4-year-olds: 6 %; adults: 25 %). This was found despite a typical explicit response, in which participants identified the box location that was in line with the false belief of the protagonist again as a function of age (3-year-olds: 13 %; adults: 95 %). This work implies that the implicit ToM system may lack the ability to differentiate several object identities and associated belief states in false-belief tracking tasks, in contrast to the explicit ToM system. This, in turn, suggests that the implicit ToM system may only track a subset of mental states, such as a protagonist's beliefs about the location of an object, but not beliefs about the object's properties.
Empirical investigations of implicit ToM processes, as described above, have also led to theorizing regarding falsebelief tracking without instructions. For example, Vierkant (2012) suggested a differentiation between completely unaware implicit false-belief understanding and aware implicit false-belief understanding. The first is described as an implicit ToM understanding that is so encapsulated that it will have only very limited influence on integrated intentional behavior. Accordingly, this type of false-belief understanding is reflected in VoE paradigms (e.g., Kovács et al., 2010;Onishi & Baillargeon, 2005;Surian et al., 2007) or false-belief anticipatory looking (e.g., Clements & Perner, 1994;Schneider, Bayliss et al., 2012;Schneider, Lam et al., 2012;Southgate, Senju, & Csibra, 2007).
In contrast, aware implicit false-belief understanding is described as implicit ToM knowledge that can guide intentional behavior, but does not figure in deliberative reports. Such implicit ToM understanding is reflected in intentional action tasks (see, e.g., Buttelmann, Carpenter, & Tomasello, 2009). These tasks are designed to reveal the participant's understanding of another person's intention, for example, to open a door, which the participant may act upon (e.g., by pulling the door handle for them). If implemented in a falsebelief scenario, participants in this type of task may not be able to explicitly report on their false-belief understanding (as in Schneider, Bayliss et al., 2012;Schneider, Lam et al., 2012); however, they typically display behavior that indicates some access to belief processing.
For example, Buttelmann et al. (2009) presented a falseand true-belief helping paradigm to 2.5-year-old, 18-monthold, and 16-month-old children. The participants watched an adult attempt to open one of two locked boxes. This occurred after a setup, which established a false-or true-belief scenario via the adult's movements in and out of the testing room. In the true-belief condition, the adult attempted to open the box that did not contain the toy. This indicated that the adult simply wanted to open the box, since the setup established that she knew the toy was located in the opposite box. In the false-belief condition, by contrast, the adult's behavior was the same but in this case she was apparently trying to access the toy, because the setup had indicated that she falsely believed the object to be in that box. The key variable here was children's helping behavior. Interestingly, 75 % of the children helped the adult open the empty box in the true-belief condition, whereas in the false-belief condition, around 83 % of children opened the opposite box, which contained the toy. The authors concluded that in the false-belief condition, children opened the opposite box because they assumed that the protagonist wanted the toy, but had a false belief as to where it was located. This suggests that such young children have an understanding of the adult's false beliefs, which they can act upon, but not explicitly report on.
For our understanding of an implicit/unconscious ToM process, the latter described conceptualization of an aware implicit ToM process fits the concept of an explicit/ conscious ToM process, simply operationalized differently (i.e., knowledge about another's mental state is processed and used in subsequent behavior). However, it remains to be investigated whether cognitively healthy participants in an intentional action task show any conscious registration of their behavior. If not, this would suggest that this paradigm also taps implicit ToM.
On that note, the same applies to the above-mentioned betting task, implemented by Ruffman et al. (2001). It may be that participants bet on certain locations without conscious knowledge of their behavior; thus, it may also be more appropriate to term this behavior an implicit ToM operation. These types of innovative belief tasks are very important in the debate about whether the implicit and explicit ToM system should be considered entirely distinct or whether they should rather be viewed as a continuum. Further, these approaches may clarify whether implicit ToM is a developmental foundation for explicit ToM, and they therefore represent one system, or whether they develop in parallel, and thus represent two systems (Apperly & Butterfill, 2009;Baillargeon et al., 2010;Perner & Roessler, 2012).
To date, only two studies have attempted to address the question of whether implicit and explicit ToM capacities are underpinned by one or two systems, and relatedly, if these capacities develop in continuation or in parallel (Clements, Rustin, & McCallum, 2000;Thoermer, Sodian, Vuori, Perst, & Kristen, 2012). Clements and colleagues reported in a training study, in a sample of over 90 children (ages 2 years and 10 months to 5 years), that only when children showed evidence of anticipatory looking in false-belief tasks, was training beneficial for improving their performance on explicit false-belief tasks. This work suggests that implicit ToM processing is a promoter/precursor for successful explicit ToM understanding. Similarly, Thoermer et al. (2012) reported, in a longitudinal study with a sample of 70 infants, that anticipatory looking in a false-belief task at 18 months significantly predicted explicit false-belief reasoning at 48 months, and this link was independent of general verbal ability. Interestingly, this was found only for matching implicit and explicit falsebelief tasks (i.e., the location-change test), but not for nonmatching implicit and explicit false-belief tasks (i.e., implicit: the location-change test, explicit: the false-content test [e.g., Smarties task]). Therefore, this might indicate that implicit and explicit ToM capacities are distinct mechanisms, which are related only when they have overlapping situational features, that is, the same task. Thus, whether success on implicit false-belief tasks is indeed a promoter/precursor for explicit ToM processing-of which false-belief reasoning is only one of many cognitive operations-remains a crucial topic of investigation. Further discussions on this point will follow later in the review.
Collectively, the above studies help shape the formulation of an implicit ToM system; however, at the same time they have motivated other investigations, which led to a range of different findings and thus brought along a variety of terms to describe implicit ToM. Labels such as "nonverbal," "nonelicited," "passive," "intuitive," "spontaneous," "social perceptual," "automatic," "bottom-up," "very simple," "minimal," and "innate" have been introduced to describe the operations thought to underlie implicit tracking of others' beliefs. In the following section we unpack these terminologies and provide an analysis of what each implies in relation to relevant psychological mechanisms. We also assess evidence for and against the involvement of such mechanisms in implicit ToM.

What processes underlie "implicit" false-belief tracking
Because false-belief tasks can be administered with both verbal (e.g., stories: Saxe & Kanwisher, 2003) and nonverbal (e.g., cartoons: Gallagher et al., 2000;animated shapes: Castelli, Frith, Happé, & Frith, 2002;Castelli, Happé, Frith, & Frith, 2000) stimuli, a body of work has investigated how these different delivery formats influence ToM. A number of different nonverbal tasks have been introduced, for example spontaneous behavior measurements (see, e.g., Scott, He, Baillargeon, & Cummins, 2012;Senju et al., 2009;Surian & Geraci, 2012), passive measurements (Geangu, Gibson, Kaduk, & Reid, 2012), and indirect reaction time measures (Kovács et al., 2010). In spontaneous and passive behavior measurements, researchers expose participants to ToM material, such as a Sally-Anne location-change task, without any instructions, and observe spontaneous behavior (e.g., looking patterns, as described above) or physiological activity (e.g., electroencephalography). In indirect reaction time measures, participants perform an unrelated task while observing different belief conditions. For example, Kovács et al. (2010) had participants detect a ball behind an occluder (sometimes present, sometimes absent) at the end of a movie sequence as quickly and as acurately as possible. The movie sequences displayed false-and true-belief scenarios; however, this was irrelevant to the participant's task. Using this setup, the authors found that the ball was detected faster when participants or protoganists in the movies correctly believed the ball to be behind the occluder than when they did not believe so.
A key motivation for the development of such nonverbal ToM paradigms is to control or eliminate the contribution of language, working memory, inhibition, and other executive operations to perform on these tasks (Hale & Tager-Flusberg, 2003;McKinnon & Moscovitch, 2007;van der Meer, Groenewold, Nolen, Pijnenborg, & Aleman, 2011). Removing the influence of these variables is particularly useful in ToM studies involving very young, healthy, or clinical populations that have limited linguistic, symbolic, or executive capacities (Samson, Apperly, Chiavarino, & Humphreys, 2004;Scott et al., 2012;Senju et al., 2009;Senju et al., 2010). ToM tests with reduced peripheral demands often show very different patterns than standard explicit ToM tasks. For example, as discussed above, children much younger than 4 years of age register the false-belief states of others (Southgate et al., 2007) but most fail explicit false-belief tasks prior to that age (Wellman et al., 2001). In addition, some researchers have tried to reduce peripheral task demands to isolate ToM processes in order to explore which brain regions are involved in processing the mental states of others (Castelli et al., 2000, Gallagher et al., 2000Geangu et al., 2012). Gallagher et al. (2000) used a story comprehension task (i.e., a version of the Strange Stories task) and a cartoon task (i.e., a task in which a cartoon was displayed, which could only be understood if an attribution of a belief state to one of the characters was made) and found that the medial prefrontal cortex activated less in the nonverbal cartoon task relative to the comprehension task.
However, it is key to note that investigations into "implicit" ToM processing that rest on nonverbal delivery formats or spontaneous, passive, or indirect response formats are not principally concerned with examining whether ToM behavior occurs outside of consciousness. Further, simply reducing peripheral task demands does not guarantee that a pure ToM operation is isolated. Therefore, these types of studies do little to disambiguate potentially distinct and overlapping operations involved in implicit/unconscious and explicit/conscious ToM processes.
Apart from these reduced peripheral demand investigations, another concept and term used in the literature to describe ToM processing in the absence of explicit instructions is stimulus-driven/bottom-up ToM (Blakemore & Decety, 2001). Based on biological motion investigations (e.g., moving dots/animated triangle paradigms: Atkinson, Dittrich, Gemmell, & Young, 2004;Castelli et al., 2002;Castelli et al., 2000;eye-gaze, hand-, and body-movements paradigms: Allison et al., 2000) this line of work considers that the perception of action serves as a prerequisite for representing another's mental state (Frith & Frith, 1999). From this viewpoint, humans even infer complex internal mental states, such as beliefs, from displays of simple twodimensional shapes, as long as the movement of the shapes is "animate" (i.e., it is self-propelled, its path is nonlinear, and it undergoes sudden changes of velocity). For example, a recent experiment by Surian and Geraci (2012), using anticipatory looking in a false-belief task, suggested that 17-month-olds show false-belief attributions to animated geometric shapes (e.g., circles and triangles). The authors concluded that even very young children could analyze the actions of unfamiliar agents to anticipate their future behavior, even in the absence of any morphological features that are typical of natural agents. Whether such nonintentional ascription of mental states to biological motion is the driving force behind a possible distinction between implicit and explicit ToM processes remains to be seen. Stimulus-driven ToM operations may allow for an implicit/unconscious ToM process to fall in place; at the same time, stimulus-driven ToM mechanisms may also be needed for explicit ToM processes (as is suggested, for example, in Leslie's model on ToM operations [1987;1994a;1994b], described in detail below).
A considerable amount of work has also been undertaken testing the umbrella term of resource-efficient ToM processing. Particularly, researchers have tried to answer to what extent individuals may "automatically" represent the mental state of another person/character/animated agent. Put differently, are inferences about beliefs, desires, and intentions made automatically when people attend to the behavior of an agent, or are such inferences made ad hoc, according to need? There is empirical support for both automatic Kovács et al., 2010;Leslie & Thaiss, 1992;Low & Watts, 2013;Onishi & Baillargeon, 2005;Qureshi, Apperly, & Samson, 2010;Sperber & Wilson, 2002;Wertz & German, 2007) and nonautomatic ToM (Apperly, Riggs, Simpson, Chiavarino, & Samson, 2006;Back & Apperly, 2010;Keysar, Lin, & Barr, 2003;Schneider et al., 2012). The first theoretical ToM model proposed by Leslie (1987;1994a;1994b) predicts that ToM processes have a foundation in a core mechanism. Specifically, it is suggested that if certain forms of agent behavior are present, this specialized ToM mechanism (ToMM) turns on reflexively and attempts to calculate the creature's mental states. As soon as ToMM has dissected the behavior of an agent to infer a set of candidate beliefs, then a separate executive processor identifies and selects the appropriate belief content from among the options. According to this theory, only with this second step is the belief ascription process complete (Leslie, Friedman, & German, 2004;Leslie, German, & Polizzi, 2005). Applying this model, one could assume that anticipatory looking behavior in implicit ToM tasks reflects this reflexively acting ToMM, whereas explicit ToM processes are the result of the executive selection processor. However, contrary to the proposal of such a reflexively/automatically operating ToMM system stands evidence from a recent study on implicit false-belief tracking in adults (Schneider, Lam et al. 2012). Schneider and colleagues used a dual-task protocol in which they had participants watch false-and true-belief scenarios while measuring their anticipatory eye-movement behavior. One group of participants just watched the belief movies (no load condition); another group, in addition to the movies, was presented with an auditory letter stream, which needed to be ignored (low-load condition), and another group counted the number of two-back letter repetitions in the letter stream and reported them at the end of the movies (high-load condition). The authors found that with the increased secondary working-memory load, implicit falsebelief belief tracking was not observed. Thus, based on established criteria for automaticity (see below), it might be that as soon as agentlike behavior is available, the ToM mechanism does not turn on reflexively or that workingmemory load disrupts this reflex. Criteria for automaticity include that the behavior and thoughts draw, at a minimum, on capacity-limited processing resources (i.e., short-term memory, attention); that they are unconscious, unintentional, and uncontrollable; that they show no benefit from practice; and that they function under all circumstances at a constant level (Bargh, 1994;Hasher & Zacks, 1979). Note that Leslie's ToM module theory does not explicitly go against the prediction that both the ToM mechanism and the selection processor may require processing resources. Thus, to unequivocally confirm whether the ToM module operates reflexively or implicit ToM mechanisms are automatic, at least the last two criteria still require investigation.
In a further effort to characterize humans' capability to track others' beliefs in the absence of being conscious of doing so, various authors have proposed names for the "implicit" ToM system; among these are "very simple" (Fodor, 1992), "social perceptual" (Tager-Flusberg, 2001), and "minimal" (Apperly & Butterfill, 2009;Butterfill & Apperly, 2013). All these proposals entail the same broad idea, mainly, that a fast, cognitively less demanding ToM system exists beside a slow and cognitively demanding ToM ability. For example, Apperly and Butterfill (2009) formulate that their minimal ToM process does not involve the representation of propositional attitudes (e.g., beliefs) but that it allows for less involved representing of mental-state-like elements. Specifically, this view suggests that the demands on perspective taking with regard to oneself or another are crucial when discriminating between effortless and efficient versus effortful and slow ToM operations. Minimal ToM operations are hypothesized to involve lower-level perspective calculations: For example, what one sees, thinks, or knows might not be seen, thought about, or known by another agent (e.g., What object/thought is available to her?). In contrast, in full ToM operations, a higher-level perspective-taking plays a key role in the judgment processes. For example, an object or thought viewed by oneself and another may nevertheless give rise to differences in what is perceived by each person about this object/thought (e.g., How is this object/thought represented by her?). Such functional differentiation of the minimal and full ToM system is consistent with results from the above-mentioned study by Low and Watts (2013). To reiterate, Low and Watts (2013) used an object with two identities, which needed to be tracked with regard to the multiple beliefs it induced for a protagonist in a Sally-Anne scenario. Implicit as well as explicit ToM behavior was measured, and only the latter reflected processing of beliefs regarding the object's properties. Thus, these findings suggested that the minimal/efficient ToM system differs from the full/slow ToM system, at least with regard to its ability to track multiple object identities.
In line with this finding, Apperly and Butterfill (2009) have proposed their minimal ToM system as a distinct faculty that supports cognitively fast, efficient, but inflexible ToM reasoning across the entire lifespan. In addition, the authors suggest that the minimal ToM system is not a developmental promoter/precursor to the full adult ToM system, but that both systems develop in parallel-particularly based on the idea that both systems demand opposing flexibility properties (i.e., the implicit system: guidance of online social interaction and communication vs. the explicit system: top-down guidance of social interaction and reasoning about the causes and justifications of mental states). Other theoretical ToM models (Baillargeon et al., 2010;Leslie, 1987Leslie, , 1994aLeslie, , 1994b, as well as empirical work (Clements et al., 2000;German & Cohen, 2012;He, Bolz, & Baillargeon, 2011;Surian & Geraci, 2012;Thoermer et al., 2012), have suggested, in contrast, a continuous trajectory of development from the implicit ToM system to the explicit ToM system. Some authors even suggest that ToM functions are an innate capacity (Leslie, 1987(Leslie, , 1994a(Leslie, , 1994bBaillargeon et al., 2010): This line of work takes the most recent implicit false-belief tracking data in very young infants (e.g., 7-month-olds, Kovács et al., 2010) as evidence for the idea that we are born with a ToM ability. In addition, it is argued that prior to the preschool period, supporting executive abilities, which would allow us to consciously represent behavior indicative of ToM understanding, are still undeveloped.
The broad differentiations between the minimal and full ToM systems as described by Apperly and Butterfill (2009;see also Perner & Roessler, 2012) fit approximately with our description of the implicit/unconscious and explicit/conscious ToM operations. However, we argue that in terms of cognitive features, implicit and explicit ToM do not represent entirely distinct mechanisms. This is predominantly based on recent studies examining the involvement of executive functions in implicit false-belief tracking behavior.
Cognitive characteristics of the implicit/unconscious and explicit/conscious ToM system As shown repeatedly in the developmental literature, ToM processes are complex and involve operations that draw, at least in their traditional explicit task formats (e.g., verbal responses to a false-belief task), on a range of general cognitive functions (Perner & Lang, 1999). Among these are inhibitory control, planning, language abilities, general intelligence, and working memory (Astington & Jenkins, 1999;Benson, Sabbagh, Carlson, & Zelazo, 2013;Hale & Tager-Flusberg, 2003;Sabbagh, Xu, Carlson, Moses, & Lee, 2006).
Many recent adult investigations have confirmed that inhibition abilities, particularly the inhibition of one's own perspective, play a crucial role when solving explicit false-belief tasks (Rothmayr et al., 2010;Samson, Apperly, Kathirgamanathan, & Humphreys, 2005;van der Meer et al., 2011;Zhang, Sha, Zheng, Ouyang, & Li, 2009). For example, Rothmayr et al. (2010) showed, using a within-subjects design, that a beliefreasoning task (i.e., a classic explicit false-and true-belief task) and a response inhibition task (i.e., the same false-and truebelief task in a go/no-go setup) activated distinct brain areas. However, at the same time, a substantial overlap for both processes was identified in the right superior dorsal medial prefrontal cortex, the right temporoparietal junction (TPJ), the dorsal part of the left TPJ, as well as the lateral prefrontal areas.
Further, it has been shown that working-memory abilities are crucial for classic ToM operations in adults (Bull, Phillips, & Conway, 2008;Dumontheil, Apperly, & Blakemore, 2010;Lin, Keysar, & Epley, 2010;McKinnon & Moscovitch, 2007;Phillips et al., 2011). For example, Lin et al. (2010) showed in a difficult explicit ToM task that under conditions of reduced working-memory capacity, people are less effective in applying their ToM abilities. Specifically, the authors employed the director communication task (Keysar, Barr, Balin, & Brauner, 2000). Here, the participant and a director sit across the table from each other, with several objects arranged in a grid of boxes between them. Some of the objects are mutually visible, whereas others are visible only to the participant. Participants are therefore conscious of the fact that the director does not know about certain objects. One situation might be that the participant sees two glasses: one glass mutually visible, and one occluded for the director. On critical trials, the director instructs the participant to move an object, but the instructions can occasionally apply to a hidden object; for example, the director could say, "move the glass up one slot," referring to the mutually visible glass. In this instance, "the glass" tends to be interpreted egocentrically; therefore the expression would be ambiguous, because the participant can see two glasses. It now requires ToM abilities to solve this task-the director's perspective has to be taken into account to identify the mutually visible target glass. Eye-gaze behavior of participants was analyzed in Lin et al. (2010) to identify which object participants would first fixate on until reaching for one. In Experiment 1 the authors found that participants with lower working-memory capacities would take a longer time from first noticing the target until finally reaching for it than participants with higher working-memory abilities. In Experiment 2, the authors found that when participants performed the director communication game along with a secondary workingmemory task (i.e., a digit-memorizing task) they took longer from fixating the target object until reaching toward it under high, relative to low, load conditions. In addition to the fact that this type of work provides compelling evidence that explicit ToM processes need the support of executive functions, such as inhibitory control and working-memory resources, it also indicates that the previously mentioned process-pure ToM operations might be very hard to validly isolate.
Along those lines, quite recently, researchers have also started to examine whether executive resources are necessary for implicit ToM operations. For example, in a study that tested 18-month-old infants, Yott and Poulin-Dubois (2012) found a strong relationship between working-memory capacities and implicit false-belief tracking abilities. As described above, Yott and Poulin-Dubois (2012) employed a VoE paradigm and found that after infants learned a new rule (i.e., an object that disappears in location A can be found in location B), they did not show looking-time evidence for this learning, but rather kept displaying the looking behavior consistent with false-belief processing (i.e., longer fixation times toward locations consistent with a violation of the expected behavior). Crucially, this effect was highly correlated with performance on a working-memory task (i.e., a detouring task on which infants had to learn that a toy should only be retrieved from behind a plastic cover after a knob was manipulated). This work converges with that of Schneider, Lam et al. (2012), who found using a dual-task paradigm that increased executive/working-memory load impaired implicit false-belief tracking (see above) while not influencing anticipatory eye movements in general. These initial results suggest that both infants' and adults' implicit false-belief tracking abilities also draw on working-memory resources.

Conclusions
A review of previous studies in both child and adult populations suggests that false-belief tracking in the absence of instructions should be described as a social analysis process that can occur without conscious registration of having engaged in such behavior. (For an overview of the defining studies in the field of implicit false-belief tracking, see Table 1.) It appears functionally similar to, but distinct from, explicit/conscious ToM, which is available to conscious processes. This distinction is supported by the research involving individuals with ASD, which has shown that explicit, but not implicit ToM, can be demonstrated in this population, and by research showing the opposite pattern in neurotypical infants and toddlers. Further support for a distinction between implicit and explicit ToM mechanisms comes from research demonstrating that, in contrast to the explicit ToM system, the implicit ToM system tracks people's beliefs about object locations but apparently fails to track people's beliefs about object identities. In sum, there is convincing evidence for the proposal of distinct ToM processes. Clements & Perner, 1994 Three-year-olds demonstrate implicit false-belief processing via eye-gaze measures but no explicit false-belief processing via verbal responding. Schneider, Bayliss et al., 2012 Implicit false-belief tracking via anticipatory looking is evident in typical adults and it is sustained over an hour of testing. Post-experimental debriefing procedure confirms that implicit false-belief tracking occurs outside adults' consciousness. Senju et al., 2009/Schneider et al., 2013 Adults with ASD demonstrate explicit false-belief processing (verbal responding) but not implicit false-belief processing (anticipatory looking), and this difference is sustained over time. Low & Watts, 2013 The implicit false-belief tracking system represents object locations but not object identities, whereas the explicit system represents both. Schneider, Lam et al., 2012 Implicit false-belief tracking is disrupted by working-memory load.
At the same time, we stress that many studies in the past did not have the core motivation of defining the cognitive mechanisms involved in implicit/unconscious mental-state processing. As described in detail above, some work has simply reduced the demands of the delivery and response format of ToM tasks. Although such work helped shape the concept of an implicit ToM system, it did not necessarily isolate the consciousness aspect of ToM. In this regard we highlighted recent work in adults, which showed that implicit ToM processes should not be considered automatic when employing classic criteria. Rather, it appears that implicit and explicit ToM processes share certain characteristics such as a demand on working-memory resources. In line with this finding, we also pointed toward past evidence in children, which indicated that implicit ToM abilities might operate as a promoter/precursor for explicit ToM functions. Thus, at this point we conclude that previous descriptions of minimal ToM operations approximately fit our concept of the implicit/unconscious ToM system; however, we believe that the description of the implicit ToM process as resource-efficient is unwarranted, given the evidence that implicit ToM processes are at least partially reliant on executive functions.
In future work we believe it would be fruitful to further shed light on what elements of information processing are crucial for complex social analysis processing. For example, to date it is still unclear what role the protagonist/s may play: Does their presence at certain times in belief scenarios trigger implicit versus explicit belief reasoning? In addition, future work on implicit ToM processes should evaluate whether described capacity limitations can be overcome with training. It will also be vital that the range of tests used in order to assess participants' ability to implicitly infer other types of mental states are expanded. Indeed, being able to represent the feelings, desires, and intentions of others, in addition to beliefs, is a key characteristic of explicit ToM (Premack & Woodruff, 1978).