Autism and Theory of Mind: A Theory in Transition

Jeramy Townsley

Autism has been a clinically diagnosed disorder since the early 1940’s, at which time Kanner and Asperger (independently) both described similar cases of children with poor social, communicative, and play skills. Kanner’s list of deficits includes (among others) the following five characteristics (Hendriks-Jansen, 1997):

1) The strong desire to be left alone, and a failure to respond to other people.

2) The obsessive desire for preservation of environmental sameness.

3) Delayed echolalia.

4) Oversensitivity to stimuli.

5) Lack of imaginative play, and perseveration of repetitive stereotypical movements (rocking, finger flicking, arm flapping).

Etiologies for autism have varied from congenital problems, genetic defects, neuroanatomical abnormalities, and poor parenting (the "cold mother"). Recent neuroimaging techniques have revealed several brain areas that show deficits, strengthening the biological case. Such areas include the cerebellum, limbic system, pre-frontal cortex, orbito-frontal cortex, as well as the opioid and serotonin systems.

However, from a cognitive perspective, the Theory of Mind (TOM) hypothesis has dominated research since the mid-1980’s. This followed from work done by Baron-Cohen, Leslie, Frith, and others, which seemed to indicate that autistic children have difficulty with second order representation. Second order representation, as explored earlier by writers such as Dennett, Premack and Pylyshyn, is the ability to impute another individual with the ability to have intentions and beliefs. For example, in a classic test of TOM, a puppet Sally, plays with a ball, puts the ball in her basket, and then leaves. Her puppet friend, Anne, removes Sally’s ball and puts it in a box. The child being tested observes all of this. The child is then asked to theorize where Sally will look for the ball. If the child guesses that Sally will look in her basket, then this is allegedly evidence that this child has second order representational capacity, since s/he believes that Sally has representational capacity (to believe that that ball is where she left it, even though the child know that ball is in Anne’s box).

At least 20 studies using this particular test, or some similar method used to determine existence of second order representational capacity support the hypothesis that autistic children lack this ability to "mind-read" (Happe, 1995). However, this dominance of support for TOM has been waning in the past few years. Several recent studies have been unable to replicate these findings. Several of these particular studies have attempted to take into account verbal ability and/or mental age matching when creating control groups. When such factors are accounted for, much less of a difference between the autistic and mentally retarded groups tend to appear than were previously reported.

This special supplemental issue represents a compilation of articles which challenge the TOM doctrine. While many theorists are not ready yet to totally abandon TOM, there is a realization that TOM leaves many phenomenon unexplained. For example, TOM does nothing to explain stereotyped and perseverative behaviors, flattened affect, restriction of interests, oversensitivity to physical stimuli and speech (prosodic) abnormalities. So while TOM may be able, on a limited cognitive level, to explain certain characteristic social and communicative features of autism, there is much more to this disorder than just a lack of second order representational capacity, if even this is lacking. The first five articles combine to support the final, meta-theoretical article, which moves away from strongly cognitive model of mental-representation to a more ethological view.

The first article, by Tager-Flusberg and Sullivan (1994), involves autistic and mentally retarded children matched for verbal mental age. In similar tests previously performed, mentally retarded (MR) individuals tended to do relatively well on TOM tests, especially when compared with autistic children. In fact, some studies showed little differences between average IQ individuals and MR individuals, with austistics performing significantly lower than both groups.

In this current study, however, such a distinction is not found. Autistics who can pass first order tests, perform as well as MR individuals on a test to determine second order capacity. The second order test used here is a simplified version of a test used by Sullivan (1994), using fewer characters, simpler language, and reduced story length.

They find that both groups respond at a level comparable to the control preschool group, indicating that both groups show second order delay, but don’t lack the capacity to represent at a secondary level. They theorize that the difficulty is not so much one of representational capacity, as information processing capacity, given that the testing procedures for second order are so much more lengthy and cumbersome than tests for first order capacity.

The second article, by Happe, comes to a somewhat similar conclusion, that verbal ability is a determining factor in passing TOM tests. Verbal age matched subjects performed with similar success on such tests. An interesting finding in Happe’s study, is that while autistics could in fact pass the second order tests, they required a higher verbal mental age than the control group to pass with similar correctness as the control group. This correlates somewhat with other data which shows that older autistics can pass TOM tests, but slowly, and with difficulty. One theory given to explain this data is that these subjects may not actually possess the ability to represent mental states of others, but are merely "hacking out" a solution to the tasks. This ability may appear by having developed compensatory strategies over time, but without fully understanding the situation. This article (and this theory) functions with the rest of these articles in this issue to support the idea that cognition may be "overrated" in the study of autism.

The third article, by Peterson and Siegal, looks at comparative rates of success on TOM tests between autistics and deaf children. They concur that verbal abilities may be a confounding variable in TOM tests. They assert that as children gain experience in conversation and social interaction, they gain the ability to properly understand references to mental representations. Thus children may very well have the capacity to have secondary representation, but misunderstand what we desire from them in our testing for TOM. In the case of the autistic, if the child shows poor verbal development, then parents and other adult figures are less likely to speak in abstract terms typically associated with mental representations. This hypothesis is supported by their research which shows little difference in TOM tests between the normal IQ-deaf children and autistics, both of whom would have received poor verbal and facial feedback as compared to infants/children with normal hearing and normal interactional capacity.

The fourth and fifth articles, by Dahlgren and Trillingsgaard, and Yirmiya and Shulman, respectively, follow along similar lines as the previous articles. When verbal intelligence is taken into account, then the results show little difference between MR groups and autistic groups. In the Dahlgren/Trillingsgaard article, differences were found in TOM tests, but only before verbal age was covaried. After this statistical factor was run, the differences dropped out. Both groups did worse than controls. Similar results were found in the Yirmiya/Schulman study, who conclude that verbal intelligence is the primary factor in determining ability to succeed in secondary representational tests, not some TOM mechanism as proposed by Leslie and Baron-Cohen.

The final article, by Hendriks-Jansen, is a provocative meta-theoretical work. He begins by giving an in-depth analysis of the computational model which underlies the TOM mechanism as provided by Leslie, followed by a critique of the capacity of this model to support the actual phenomenon of autism and mind-type behaviors. The first component Leslie’s model is a decoupler, which separates the original, experienced representation from any subsequent, fantasy or experimental representations. The next two components are a manipulator, to transform the decoupled representation into pretend representations, and an interpreter, to relate the pretend representation to the situation at hand. The final component that had to be added to counter certain criticisms, is a selection processor, which handles knowledge about the represented artifacts. These four components allegedly allow us to have a TOM. This mechanism is entirely nativist, and relies totally on the innate ability of the developing infant to process all aspects of his/her world in this orderly, complex fashion.

Hendriks-Jansen, however, proposes a mechanism based more in chaos theory, relying on the biological principle of epigenesis. Epigenesis "conceptualizes development as a cumulative process," and that "species-typical development can proceed only if the partly completed phenotype has developed appropriately thus far, the required gene products are available, and the environmental conditions fall within a certain range" (378-88). His theory closely resembles a rigid behaviorist view from decades ago, incorporating small steps in the production of behavior that eventually appears to exhibit mental representations.

He relies on his experience with robotics and AI programming to support his theory. Similar to the fact that simple robots can exhibit primitive meaningful behavior without having been programmed with internal representational capacity, so might humans be able to exhibit more complex behaviors without internal representation. He points to ethological data on species-typical behaviors that support the idea that infants of a species engage in behaviors which exhibit certain behavioral patterns from their care-givers. Their care-givers, in turn, engage in behaviors which produce survival-specific behaviors from their infants. Similarly, human mothers, when interacting with their babies (who engage in species-typical behaviors that tend to produce meaningful responses in adult humans), are primed to impute their babies behaviors as having intention and beliefs. Hendriks-Jansen goes to great lengths to show how this constant, dynamic interaction between mother and child forms a "scaffolding" which supports behaviors appearing mentalistic, so that at a later time, when the reinforcing scaffolding disappears, the mentalistic behaviors still remain.

In the end, he paints a picture of how humans develop into appearing to have mentalistic behaviors. It is defects in this early interaction process between parent and child that produce maladaptive behaviors of later development. This isn’t an shift back towards blaming the parent. Rather, it is an attempt to show that minor defects in early neurological development prevent the infant from producing species-typical behaviors (smiling, cooing, pointing, etc.), prevent normal interactions between parent and child. It is, according to Hendriks-Jansen the lack of normal verbal and para-linguistic communication to the infant and child that produce the poor social skills and meta-representational capacities of the older child. This development is not entirely stopped, but is dramatically slowed down, and heuristics different than the average human may develop, as is theorized in the earlier articles by Happe, and Tager-Flusberg.