Representing metarepresentations: Is there Theory of Mind-specific cognition?

Abstract

What cognitive mechanisms underlie Theory of Mind? Some infer domain-specific Theory of Mind cognition based the pattern of children diagnosed with autism failing the False Belief test but passing the False Photograph test. However, we argue that the False Belief test entails various task demands the False Photograph task does not, including the necessity to represent a higher-order representation (a metarepresentation), thus confounding the inference of domain-specificity. Instead, a general difficulty that affects representations of metarepresentations might account for the seeming domain-specific failure. Here we find that False-Belief failing False-Photograph passing children fail the Meta Photograph test, a new photograph-domain test that requires subjects to represent a metarepresentation. We conclude that people who fail the False Belief test but pass the False Photograph test do not necessarily have a content-specific Theory of Mind deficit. Instead, the general ability to represent representations and metarepresentations might underlie Theory of Mind.

Introduction

People remember what they have seen, and people think about what other people think (about what yet other people think): higher-level representations, or metarepresentations, are ubiquitous in mental life, and the cognitive ability to represent metarepresentations is essential to understanding our own and other people’s psychology. A prominent theory holds that we think about mental states by means of a domain-specific “Theory of Mind Module,” a dissociable piece of cognition that specifically represents mental representations like beliefs, but not non-mental representations like photographs or logical propositions (Leslie, 1994). The theory of a Theory of Mind Module is opposed to the theory that representations of both mental and non-mental representations are created by the same cognitive processes (e.g. Zaitchik, 1990). Here, we question the inference that evidence from patient populations and neuroimaging supports a Theory of Mind Module in human psychology, and we suggest instead that the general ability to represent representations and metarepresentations underlies the human ability to understand our own and others’ minds.

The primary evidence for the representation of minds by Theory of Mind-specific cognition comes from two findings based on the False Belief and “False” Photograph tests¹. First, children diagnosed with autism fail the False Belief test, which requires subjects to represent mental representations, but they pass the “False” Photograph test, which requires subjects to represent non-mental representations (Leekam and Perner, 1991, Leslie and Thaiss, 1992). Second, the brains of healthy adults show more activation in the temporo-parietal junction during the False Belief test compared to the “False” Photograph test (Saxe & Kanwisher, 2003).

In the False Belief and “False” Photograph tests, an object is changed after a character leaves the room or a Polaroid camera takes a picture (respectively), and to pass the test, subjects must demonstrate that they understand that the representation in question, either the character’s memory or the Polaroid photograph, does not track the change. Because both tests require subjects to represent representations, but only the False Belief test includes mental representations, some reason that selective False Belief failure in autism implies a content-specific Theory of Mind deficit in autism and an intact content-specific Theory of Mind module in typical development and, furthermore, that brain activation differences during these two tasks indicates a physical location for the module (e.g. Baron-Cohen, 1995, Leslie, 1994, Saxe and Kanwisher, 2003).

However, a closer examination of the False Belief and “False” Photograph tests reveals that to represent the mental representations in the False Belief test entails more cognitive demands than to represent the non-mental representations in the “False” Photograph test, meaning that selective False Belief test failure might be due either to the mental-state content of the False Belief test or to the higher demands of the False Belief test. Some such demands others have identified include working memory load (Davis and Pratt, 1995, Gordon and Olson, 1998), the need to resist interference from reality (Apperly, Samson, & Humphreys, 2005), the need to reconcile a conflict between one’s own beliefs and a character’s beliefs (Russell, Saltmarsh, & Hill, 1999), the lack of straightforwardness of the causal history of the belief representation (Muller, Zelazo, & Imrisek, 2005), and the need to update beliefs versus the fixedness of photographs (Sabbagh, Moses, & Shiverick, 2006). As we explain below, the lack of parallel task demands between the False Belief and “False” Photograph tests was intended by the creator of the “False” photograph test (Zaitchik, 1990), but makes these tests inappropriate to use to infer a domain-specific Theory of Mind deficit. Later, we identify an additional key task demand difference: the False Belief test might entail more levels of representation than the “False” Photograph test.

Zaitchik (1990) originally found that typically developing children who fail the False Belief test also fail the “False” Photograph test, concluding that the ability to represent domain-general representations might underlie belief representations. The association on test performance observed by Zaitchik (1990) suggests that domain-general cognition represents mental-state representations, even though—especially because—various task demand differences make the False Belief test more demanding. In comparing the two tests, Zaitchik (1990) states,

Perception, after all, is a complicated process; it does not occur in a discrete moment and it rarely calls attention to itself. The representations it engenders—beliefs—are immaterial and abstract. In the case of the photograph, however, it’s hard to see the same argument being made; in this condition, the process of fixing the representation, focusing on an object and pushing the button, has none of the complexity of perception. Unlike the mind, the operation of the camera is behaviorally salient and temporally discrete. Furthermore, the representations themselves, the photographs, can be seen and handled...

Consider the photographs themselves; none of the properties of beliefs which have been taken to cause the child’s problems apply to the photos. Photos are not immaterial, they are not intangible, they are not private and internal … nevertheless, [children] fail. (p. 62).

Because the “False” Photograph test is less demanding, the conclusion becomes clearer that failure on both it and on the False Belief test is due to the requirement for subjects to represent a representation per se rather than any of the other cognitively demanding difficulties associated with the False Belief test. However, selective failure on the False Belief test—a dissociation in test performance—would not indicate a Theory of Mind-specific deficit because failure might be due to any of the extra demands of the False Belief test. Therefore, contra Leslie and Thaiss (1992), we conclude that we do not know if False-Belief failing “False” Photograph-passing children have a domain-specific Theory of Mind deficit: they might, or they might just as well have a deficit that affects their ability to process any of the domain-general task demands inherent to the False Belief test that Zaitchik (1990) intentionally left out of the “False” Photograph test. The inherent domain-general task demand differences between tests are especially problematic in the study of possible domain-specific deficits in autism, which is known to include domain-general executive deficits (e.g. Perner & Lang, 1999): children diagnosed with autism would be expected to have less success with the more demanding test (the False Belief test) separate from any question of a domain-specific deficit (see Stone and Gerrans, 2006a, Stone and Gerrans, 2006b).

Even children diagnosed with autism who fail the False Belief test can typically represent some Theory of Mind representations, such as other people’s perception (e.g. Baron-Cohen and Goodhart, 1994, Hobson, 1984). Premack and Woodruff (1978) define a Theory of Mind as that which separates mentalists from behaviorists, the term applying equally to mentalizing academics’ explicit theories about mental states and mentalizing chimpanzees’ implicit theories that guide their behavior with respect to predicting others’ behavior (their devious proposition is that behavorist theories cannot fully explain chimpanzee psychology if the chimpanzees themselves are mentalists!) A behaviorist has no more use for what a person sees than for what a person knows, so, if children diagnosed with autism represent other people’s perceptions, then they have Theories of Mind. By design, passing the False Belief test requires a Theory of Mind (Wimmer & Perner, 1983), but failing the test does not rule out a Theory of Mind.

Likewise, passing the “False” Photograph test does not rule out a domain-general deficit that affects the ability to represent representations. The Polaroid photographs in the “False” Photograph test are, by elegant experimental design, as undemanding non-mental-state representations as possible, as Zaitchik (1990) was exploring the theory that the mere requirement to represent a representation would cause some individuals to fail a test. Any representational deficit that does not entirely eradicate the ability to represent all representations might preserve “False” Photograph performance. A domain-general representational deficit might let a subject understand the first-order photograph representations in the “False” Photograph test but fail to understand the more difficult belief representations in the False Belief test. Below, we argue that beliefs in the False Belief test might be higher-order representations than photographs in the “False” Photograph test; if so, False-Belief failing “False” Photograph passers could just as well have a domain-general representational deficit as a Theory of Mind-specific deficit.

The content of a characters’ belief in the False Belief test is based on the content of that character’s prior perception, making the belief a higher-level representation than the perception. Zaitchik (1990) notes that the False Belief test requires subjects to represent complex perceptual processes as well as subsequent post-perceptual belief representations. Dawkins (1986, p. 15) and Sekular and Blake (1985, p. 35) point out the striking analogy between the photographic representations created by cameras and the visual perception of eyes: both have a lens, retina/film, and a resulting image/percept that reflects the visual information in a scene. Photographs are long-lasting like beliefs, but as far as representational complexity, nothing in the photographs of the “False” Photograph test goes beyond the analogy of visual perception and seeing, whereas in the False Belief test, the contents of characters’ beliefs are based on what those characters previously saw, making these beliefs higher-order representations (metarepresentations). If the False Belief test entails more or higher-level representations than the “False” Photograph test, then the degree of a domain-general representation deficit, rather than the type of deficit, might account for the difference between a False-Belief failing “False” Photograph failure and a False-Belief failing “False” Photograph passer.

An individual who has trouble representing metarepresentations might be unable to understand higher-order photographs, meta-sentences, post-perceptual beliefs, and a wide range of higher-order social communication such as common knowledge (ego knowing that other knows that ego knows that other knows, etc. [Schelling, 1960]), human contact, eye contact, and joint attention (ego representing other representing ego, etc). Self-knowledge and self-awareness might also be impaired, which some models predict would lead to decreased empathy (e.g. Gallup, 1998). Self-referential statements entail metarepresentations, and while some are confined to logic and math, they are also common in social communication: for example, any time a speaker uses the word “I,” the listener must represent that the entity which is doing the referring (“I”) is the same as the entity which is being referred to (also “I”; see Hofstadter, 1979, Legrand, 2007, Zahavi, 2007).

There are multiple lines of evidence that point to a deficit in representing metarepresentation in autism. Children diagnosed with autism who pass the False Belief test, unlike children diagnosed with Down Syndrome who pass the False Belief test, go on to fail a meta-belief test (beliefs about beliefs, Baron-Cohen, 1989). Autism includes abnormal self-referential cognition including self-recognition (Dawson & McKissick, 1984) and self-related memory (Klein et al., 1999, Millward et al., 2000). Children diagnosed with autism fare worse on tests of irony than tests of metaphor; both are forms of non-literal language, but only irony requires representing metarepresentations (Happé, 1995). Diagnosed children also show poor communication, which requires establishing common knowledge (which, in turn, entails representing metarepresentations [Sperber & Wilson, 1986]), as well as poor joint attention and eye contact (e.g. Neumann, Spezio, Piven, & Adolphs, 2006), other instances where individuals must represent other individuals who are in turn representing representations. In addition, Neibauer and Garvey (2004) link poor understanding of Godel’s logical paradox and Escher’s self-referential prints, both examples of non-mental-state metarepresentations, to a small corpus callosum, a prominent feature of brains of children diagnosed with autism (Hughes, 2007).

Autistic children seem to know that “cookie” means cookie and “broccoli” means broccoli, and that one cannot eat the word “cake:” they have a basic understanding of reference, representation, and language use. But, they have more difficulty with the self-referential statement that “I” want a cookie. Children with autism, like typically developing precocious talkers, make “I” vs. “you” pronoun reversals (Szatmari, Bartolucci, & Bremner, 1989). Language and representation is not the problem here, but higher-order language and metarepresentation might be. In this vein, Tager-Flusberg and Joseph (2005) find that understanding recursive embedded sentences predicts False Belief success. Similarly, Zelazo, Jacques, Burack, and Frye (2002) find that False-Belief failure in autism is associated with failure to understand higher-order rules (rules about rules) that have nothing to do with mental states.

The observed dissociation on False Belief and “False” Photograph task performance in autism but not typical development tells us that some task demand of the False Belief test overlaps with the deficit in autism, whether that demand be the mental content per se or one or more of the representational and non-representational confounds described above. The deficit in autism can tell us something about the cognitive basis of typical human Theory of Mind, as a Theory of Mind deficit (specific or otherwise) is a salient feature of the disorder. Our basic interest is in dissecting the cognitive architecture of Theory of Mind, which turns not on a diagnosis per se but rather on the pattern of performance on the tasks in question. That is, our interest here is not in autism in and of itself, but rather in selective deficits in cognition as indicated by passage and failure on the tasks in question. Autism is an interesting case, but False-Belief failers, especially those who pass the “False” Photograph test, are generally interesting: do they have a Theory of Mind-specific deficit or a general inability to represent metarepresentations? In this way, probing the cognitive deficit of False-Belief failing “False” Photograph passers can be very telling for the study of typical development.

Therefore, in order to determine whether False-Belief failing “False” Photograph-passers might have a domain-general deficit that affects representations of higher-order representations, we introduce the Meta Photograph test, which requires subjects to represent a photograph of a photograph. We also introduce a non-meta control photograph test, the Strange Blindness test, so-called due to its similarity to change blindness paradigms (Simons & Rensink, 2005), which entails equal or greater task demands as the Meta Photograph test. Our study finds a set of individuals with the cognitive phenotype said to evidence a domain-specific Theory of Mind deficit, namely, False-Belief failing “False” Photograph passers, and asks this question: Does their deficit extend to higher-level photographs? If so, they have representational deficits that are not specific to the domain of Theory of Mind. If not, then we would be a step closer to establishing that some individuals have a Theory of Mind-specific deficit rather than an inability to process the domain-general task demands of the False Belief test.