Herbert Simon says "there may be no such gulf between the meanings of the author and the meanings of the text as is presumed in some contemporary theories of criticism"(18). I believe that recent studies of cognitive development support the authorial/textual divide. If there is disagreement with Simon here it should not obscure my enthusiasm for his basic message: the potential of a cognitive approach to literary theory (see also Currie, 1994). I hope very briefly to underline that message by showing how natural, as well as artificial, intelligence can suggest mechanisms of meaning.
Central to the condition of autism is a marked failure to engage with the mental lives of others; autistic children typically treat others, including caregivers, as "objects." We now know that autistic children fail to make an important cognitive leap that usually takes place around age four when children learn that others may have different beliefs about the world from their own (Baron-Cohen et. al., 1985). Remarkably, this deficit seems to be quite localized. While autistic children usually fail the elementary "false-belief" tests designed to assess understanding of mental states and almost always fail more complex "higher order" tests on beliefs about beliefs, their comprehension of behavioral and mechanical processes is good. (The debate here focuses on the 20% of autistic individuals who don't also suffer general mental retardation.) There is also a growing perception that the communicative difficulties of autistic children reflect a problem with the pragmatics of language, rather than a specifically semantic deficit (Tager-Flusberg, 1993).
Now consider again the hypothesis that there is a psychologically real distinction between authorial and textual meaning: what I shall call the reality thesis. Let us elaborate the thesis a bit: readers apply purely semantic competence to grasp the textual or conventional meaning of discourse but need access to utterer's intentions in order to grasp nonliteral tropes like metaphor and irony. On that hypothesis, subjects with a known disability in understanding mental states but normal comprehension in other areas would do well on understanding textual meaning and poorly on comprehension of metaphor and irony. A preliminary test of that prediction has now been made (Happa, 1993).
In one experiment, autistic subjects were asked to complete a metaphor and a simile, the point being that a simile can be understood in terms of its literal meaning, while a metaphor cannot. Subjects were presented with a number of incomplete sentences, such as "The dog was so wet, it was like. . . .," and some alternative completions: "a brick wall," "diamonds," "a walking puddle." One would not expect autistic performance to be perfect, since the appropriateness of judgements about likeness depends partly on one's purpose, and picking out intuitively the "right" simile might depend to some extent on perception of the utterer's intention. Still, a group of autistic children, chosen because they failed the elementary false-belief test, got the answer right three out of five times. When asked to make a comparable choice for metaphor ("Michael was so cold. His nose really was. . . .": "an icicle," "a fox," "a hat"), their performance was only half as good. On a task that seems likely to call only on competence with textual meaning-choosing a synonym-they were successful nearly four times out of five. The more able group who passed the elementary false-belief test but failed its higher-order variant did extremely well on the simile test and just slightly worse on the metaphor test. The third and most able group who passed even the higher-order false-belief test performed near ceiling on both the simile and the metaphor task. Thus ability to comprehend the mental states of others, as measured independently by the false-belief tests, correlates well with ability to understand metaphor, confirming the hypothesis that nonliteral discourse is processed by appeal to utterer's intention in a way that literal discourse is not.
The same three groups were compared for comprehension of metaphor and irony. What is especially interesting is that the results indicate that ability to comprehend mental states was a stronger factor in determining success in the case of irony than in the case of metaphor. Why? Happa notes that there may be evidence here for the view that an ironic utterance consists of quoting a hypothetical thought ("What a lovely shirt") and expressing an attitude of mockery towards it (Sperber and Wilson, 1986). Comprehension of irony would then require a more sophisticated insight into the utterer's mind: grasp of a thought about a thought. Sperber and Wilson's theory predicts that while understanding metaphor requires only success on the elementary false-belief test, understanding irony requires passing the higher-order variant. In broadly statistical terms, the prediction was confirmed.
Happa's results need to be replicated under a variety of conditions before they can be regarded as robust. Nevertheless, they give prima facie support to the reality thesis. They certainly indicate one way to connect literary theory with a science of the mind rich in testable consequences.