Journal of Autism and Developmental Disorders, VoL 18, No. 3, 1988
Social and Pragmatic Deficits in Autism:
Cognitive or Affective? 1
Simon Baron-Cohen 2
University College, London and St. Mary's Hospital Medical School, London.
Autism is characterized by a chronic, severe impairment in social relations.
Recent studies of language in autism also show pervasive deficits in prag-
matics. We assume, uncontroversially, that these two deficits are linked, since
pragmatics is part of social competence. This paper reviews the literature
describing these deficits, and then considers two different psychological the-
ories of these phenomena: the Affective theory and the Cognitive theory.
Although the Affective theory makes better sense of the results from emo-
tional recognition tasks, the Cognitive theory predicts the particular pattern
of impaired and unimpaired social skills in autism, as well as the pragmatic
deficits. These two theories might usefully be integrated in the future.
Autism is characterized by a chronic impairment in social relations (Kanner,
1943). This is a necessary feature in all systems used in the diagnosis of au-
tism (e.g., DSM-III, American Psychiatric Association, 1980; ICD-9, World
Health Organization, 1978), and is widely seen as the primary symptom (Fein,
Pennington, Markowitz, Braverman, & Waterhouse, 1986; Rutter, 1983).
Speaking autistic children are also impaired in the pragmatic aspects of their
language (Cromer, 1981; Tager-Flusberg, 1981, 1985). This feature does not
appear in the diagnostic systems (despite the fact that no cases of speaking
autistic children with normal pragmatic competence have ever been report-
ed). The definition of pragmatics is using speech and gesture in a communica-
1The ideas described in this paper were first presented at the symposium on Semantic and
Pragmatic Disorders at Newcastle University in April 1986, and at the BPS Developmental
Psychology Conference, Exeter, UK, in September 1986. I am grateful to Peter Hobson
and Pat Howlin for their comments on earlier drafts of this paper, and to the reviewers for use-
"Address all correspondence, including requests for reprints, to Simon Baron-Cohen,
Psychology Department, Institute of Psychiatry, Decrespigny Park, London, SES 8AF, UF.
0162-3257/88/09000379506.00/0 y 1988 Plenum Publishing Corporation
tive way, appropriate to the social context (Bates, 1976). In other words,
pragmatic skills are defined as a part of social skills. It is reasonable to as-
sume, then, that whatever underlies the deficit in social skills is also likely
to underly the deficit in pragmatic skills. In this paper we review the litera-
ture concerning the social and pragmatic deficits in autism, and then con-
sider two different theories concerning their underlying psychological basis.
(In reviewing these deficits, social and pragmatic skills are considered
separately, but it follows from our earlier assumption that no theoretical claim
is implied by this separation.)
SOCIAL DEFICITS IN AUTISM
Reviews of the literature on social deficits in autism have appeared re-
cently (Howlin, 1986; Shah & Wing, 1986; Stone & LaGreca, 1986; Volk-
mar, 1987), so the relevant studies are only summarized here.
Studies of Social Behavior
The earliest descriptions of the social impairment in autism are by Kan-
her (1943) and by Kanner and Eisenberg (1956). These take the form of clin-
ical impressions. Their papers contain references to at least 12 different
aspects of social impairment among the 11 cases described. These include
lack of "apparent affection" (p. 2), withdrawal from people (p. 2), lack of
attention to people (p. 32), noncommunicative use of language (p. 27), lack
of communicative gestures (p. 8), treating parts of people as detached ob-
jects (p. 27), lack of eye contact (p. 26), treating people as inanimate objects
(p. 15), lack of behavior appropriate to cultural norms (p. 30), attention to
the nonsocial aspects of people (p. 31), lack of awareness of the feelings of
to others (p. 95), and lack of savoire-faire (p. 94). (Page numbers refer to
the edition of collected papers, Kanner, 1973.) Most of these observations
have been supported and refined by later studies.
Wing and Gould's (1979) epidemiological survey of handicapped chil-
dren in the London borough of Camberwell revealed that social impairment
is not restricted to autism but is also found among other mentally handicapped
people. They found that 21.2 of every 10,000 children aged under 15 years
in the area showed impairments of reciprocal interaction and, of these, 4.9
had a history of typical autism. Furthermore, they found that the social im-
pairment could be distinguished into three types: social aloofness, passive
interaction, and active-but-odd interaction. This latter description referred
to social behavior that was undertaken mainly to indulge some repetitive,
idiosyncratic preoccupation, showing no interest in the other person's needs.
Social and Pragmatic Deficits in Autism
Their study highlighted that not all autistic children show withdrawn, aloof
social behavior (although 70~ of their autistic subjects fell into rids category),
and that many do indeed approach and attempt to interact with others, but
in inappropriate ways. Wing (1978) reported that the full, classic picture of
aloofness and detachment seems much more marked in the younger autistic
child, of less than 5 years old. Many authors make the point that one should
be careful not to interpret any remittance of withdrawal seen in older autis-
tic children as the onset of normal social behavior, since it is possibly only
a sign of shifting between the categories of social impairment (Frith, 1982).
Hopkins and Lord (1981) found Wing and Gould's three categories of
social impairment were useful descriptions for rating autistic children, and
could be measured in terms of the number of initiations and responses to
interactions. They found that the category that any one autistic child fell
into varied according to the age, sex, familiarity, and diagnosis of the play-
mate. They concluded that an autistic child's social impairment thus takes
different forms according to whom they are with, but it nevertheless per-
sists. The impairment also changes developmentally. Lord (1984) proposed
a progression from "aloof" to "passive" in social responsiveness, and from
"aloof" to "passive" to "odd" in rate of initiation of interaction.
Hopkins and Lord's (1981) study showed clearly that autistic children
do take account of other people's behavior. This was also found by Sussman
and Sklar (1969) and Clark and Rutter (1981) who found differential social
responsiveness to varied tone of voice and amount of interpersonal demands,
respectively. These latter two studies measured social behavior in terms of
degree of compliance, and this is obviously only a small part of social skills.
Few papers have given much space to discussion of the definition of social
behavior, and this has led to rather crude measures being used. For exam-
ple, McHale (1983) scored children as part of a group "if they were judged
to be within 5 feet of one another, or were playing on or with the same toy"
(p. 87). Clearly however, neither physical proximity nor action on someone
else's toy necessarily involves social behavior. The definition of what consti-
tutes social behavior requires a separate paper in itself and cannot be dis-
cussed here, but it is worth noting that, in the literature on normal child
development, one way in which social behavior has been discussed more
thoroughly is in terms of "mutually intentional relations" (Damon, 1979; Frye,
1981). This approach has recently been applied to autism by Mundy, Sig-
man, Ungerer, and Sherman (1986), Sigman, Mundy, Sherman, and Un-
gerer (1986), and Loveland and Landry (!986) who found that autistic children
showed significantly less "joint attention" than matched controls, and
"showed" or pointed to toys less often.
Dewey and Everard (1974) reported that social abnormalities such as
nonreciprocal speech are evident even in autistic adolescents of normal in-
telligence. Persisting social difficulties in able autistic adults were also found
by Newson, Dawson, and Everard (1984). Dewey and Everard's subjects were
also unaware of such dimensions as social class and social status in others.
These observations merit further experimental investigation. Rutter, Green-
field, and Lockyer (1967), in their follow-up of the social outcome of 63 au-
tistic children, documented a very poor prognosis, and Rumsey, Rapoport,
and Sceery (1985) confirmed this picture. The social impairment thus ap-
pears to be lifelong. Even Gajzago and Prior's (1974) description of two peo-
ple who had "recovered" from autism showed clear and persisting social
It is important to stress, however, that autistic children's social develop-
ment is not impaired in a blanket fashion. For example, Sigman and Un-
gerer (1984) found that autistic children do show some attachment behavior
(e.g., proximity-seeking) after reunion with their care-giver, and Mundy et
al. (1986) and Sigman et al. (1986) found simpler levels of social interaction
(such as eye contact and reaching after tickling) were present in 3- to 6-year
old autistic children. Similarly, Wetherby and Prutting (1984) found autistic
children do exhibit gestural requests for social routines, and Curcio's (1978)
study confirmed that requesting toys using gestures ('protoimperatives') is
also within their ability. Areas of unimpaired social functioning have also
been documented in experimental studies of autistic children's social under-
standing, described in the next section, although these studies indicate se-
vere impairments in specific areas as well.
Experiments in Social Understanding
Hutt and Ounsted (1966) investigated the phenomenon of "eye-gaze
avoidance." They found autistic children looked at people's faces less than
controls, and this has also been found by Richer (1976) and Castell (1970).
This result was refuted, however, by O'Connor and Hermelin (1967) who
found that autistic children simply have shorter, more frequent fixations for
all types of stimuli, and not faces in particular. Their finding was replicated
by Davids (1974) and Langdell (1981). O'Connor and Hermelin also found
that both autistic and normal children spent more time looking at a real face
than at a photographed face, and spent equal amounts of time looking at
a face with its eyes open or shut. On the basis of these results, O'Connor
and Hermelin seriously questioned the very existence of the phenomenon of
eye-gaze avoidance in autism.
Nevertheless, the matter remains controversial, as Richer and Coss
(1976) reported evidence apparently refuting O'Connor and Hermelin's (I 967)
results. All of these studies may, however, be missing the social use of eye
gaze by focusing on the quantitative aspects. Mirenda, Donnellan, and Yo-
der (1983), in a pilot investigation, found qualitative differences between
Social and Pragmatic Deficits in Autism
eye gaze use in autistic and normal children: Autistic children tended to look
for longer periods of time and more frequently during monologues than did
normal children. This abnormality may be related to deficits in turn-taking
in dialogue (discussed later), in which eye signals play an important part (Ar-
In tests of face-recognition, Langdell (1978) found autistic children
were able to recognize their peers in photographs, and found they made fewer
errors than their controls when the lower half of the face was shown only.
This suggests they were less dependent on the information contained in the
upper parts of the face, perhaps the eye region, for recognition. In addition,
the older autistic chidren were better than their controls at recognizing the
face when it was inverted, although they too showed the well-known inver-
sion effect. This suggests they may use a qualitatively different strategy in
face-recognition. Goode's (1985) finding that autistic adults are superior to
controls at recognizing faces from achromatic photographic negatives sup-
ports this view. This area clearly merits further investigation.
In testing comprehension of emotions, Hobson (1986a, 1986b) found
that autistic subjects made more errors in choosing schematic faces to match
videotapes showing emotions expressed in gesture, vocalization, or context.
This intermodal matching of different emotional indices appears to be more
difficult for autistic subjects than recognizing emotions in one modality:
Langdell (1981) found that autistic children were able to sort photographs
of different emotional expressions (in faces) significantly above chance, and
Weeks and Hobson (1987) found 6 out of 15 children spontaneously sorted
photographs by facial expression, and 4 others did so when told that this
was the dimension of interest. Jennings (1973) found that autistic children
prefer to sort photographs of faces according to nonaffective stimuli (e.g.,
hats) rather than expressions, unlike matched controls. Weeks and Hobson's
study obtained similar results.
In other experiments, Hobson (1983, 1987) investigated autistic chil-
dren's recognition of age and sex, and found impairment. However Weeks
and Hobson (1987) and Abelson (1981) found no impairment in sex recog-
In tests of visual self-recognition, autistic children are unimpaired, as
shown in their understanding of their own reflection in mirrors (Flannery,
1976; Neuman & Hill, 1978; Ferrari & Matthews, 1983; Spiker & Ricks, 1984;
Dawson & McKissick, 1984; Baron-Cohen, 1985). They thus appear to have
a concept of self, as an object of their own perception. These tasks of course
measure only one aspect of the concept of self (i.e., as a separate physical
object) and impairment may exist at other levels. For example, three of the
mirror studies reported a striking lack of shyness, embarrassment, or coy-
ness in front of the mirror (Baron-Cohen, 1985; Neuman & Hill, 1978; Spiker
& Ricks, 1984), and such self-conscious reactions are found in normal chil-
dren (Amsterdam & Greenberg, 1977; Dixon, 1957) and in Down's syndrome
children (Marts, Cicchetti, & Sroufe, 1978).
In another study, Hobson (1984) found that autistic children were unim-
paired in their perceptual role-taking ability in three different tasks. Baron-
Cohen (1985) also found perfect performance by autistic children in know-
ing what another person was looking at. However, significant differences
between autistic and control children were found on tests of conceptual role-
taking (Baron-Cohen, Leslie, & Frith, 1985), in which autistic children were
impaired in their ability to predict where a person would look for an object
if it was moved from its last location in the person's absence. This result was
seen as a failure to attribute different beliefs to others, or to use a "theory
of mind" (Premack & Woodruff, 1978). Similar results have been obtained
using a picture-sequencing paradigm in which autistic children's ability to
sequence social stories depended on whether attribution of mental states to
the characters was required (Baron-Cohen et ai., 1986). A similar picture
has also been obtained using a gift-choosing paradigm (Dawson & Fernald,
A number of authors have investigated autistic children's imitation abil-
ity. DeMyer et al. (1972) found that, in autism, imitation of body move-
ments was at a lower level than imitation of object use. Dawson and Adams
(1984) found that very few of their autistic children showed Stage 6 perfor-
mance of imitation using the Uzgiris and Hunt (1975) scale, but most were
in the retarded IQ range, and the subjects who were at the ceiling on the
imitation scale had a higher verbal mental age. Van Smeerdjik (1981) also
found imitation was related to mental age. Jones and Prior (1985) found
that imitation at lower levels was unimpaired, but was impaired at higher
levels. Curcio (1978) and Hammes and Langdell (1981) have confirmed that
imitation per se is not an autism-specific deficit (as indeed autistic children's
excellent echolalia testifies), but imitation of abstract gestures is difficult for
autistic children. Bartak, Rutter, and Cox (1975) and Ohm (1987) found a simi-
lar picture. The possibility has been raised that this reflects a form of dys-
praxia (Jones & Prior, 1985).
To summarize, studies in autistic children's social behavior document
the chronic nature of the social deficit, and suggest that although it may
change its form both developmentally and across situations, an inability to
participate in two-way reciprocal social interaction persists throughout the
lives of autistic people. The studies of autistic children's social understand-
ing have shown a number of unimpaired areas, such as face recognition, mir-
ror self-recognition, and perceptual role-taking, but severe impairments have
been found in intermodal matching of emotional expressions, in conceptual
role-taking, specifically in attributing different beliefs to others, and in imi-
tation of symbolic gestures. An important task for psychological theories
in this area is to account for why autistic children's social understanding as-
Social and Pragmatic Deficits in Autism 385
sumes this particular uneven profile. We return to this question after reviewing
the pragmatic deficits that have been found in autism.
PRAGMATIC DEFICITS IN AUTISM
There is no single study that has comprehensively assessed pragmatic
skills in the language of autistic people, but many studies have looked at
isolated aspects of pragmatics in autism. Some of the early studies we review
do not use the term pragmatics but nevertheless describe deficits that would
be covered by that term today. For example, Kanner (1943) spoke of a failure
to use speech for communicating meaning to others, and Rutter et al. (1967)
noted that those autistic children who developed language showed a tenden-
cy towards obsessive questioning. Both of these observations fall into the
domain of pragmatic deficits.
Another early study of autistic children's language (Cunningham, 1968)
divided speech into "egocentric" and "socialized", and found more egocen-
tric remarks in autistic children's speech than in matched controls. Egocen-
tric speech comprised echolalia, self-repetition, thinking aloud, and
apparently purposeless remarks. Cunningham did not find autistic children
asked more questions than control children, but did find their questions relat-
ed more to obsessional interests. He also found autistic children made far
fewer remarks giving spontaneous information. He discussed the excess of
egocentric speech in terms of Piaget's (1932) theory of young normal chU-
dren's egocentrism, and concluded that autism may represent an immaturity
of development. Cunningham wrote:
As Piaget (1932) points out, the exchange of information requires the speaker to place
himself at the point of view of his hearer. This the psychotic lie: autistic] child is
unable to do. He shows a lack of empathy or ability to apprehend his hearer's state
of mind and therefore falls back on non-communicative or demanding speech. (p. 243)
In an early case description of an autistic child's language (Shapiro, Fish,
& Ginsberg, 1972) the noncommunicative use of speech was again stressed,
despite normal language test results. This was particularly seen in terms of
a lack of sharing of information.
The first study to specifically examine pragmatic skills in autism was by
Baltaxe (1977), who compared autistic adolescents' discourse to that of the
normal children studied by Keenan and Klein (1975). She found that that
autistic children frequently failed to shift out of the hearer role to become
a speaker. Baltaxe (1977) quoted one of her autistic subjects: "Well, I asked
my parents. I told my parents I'd be good at home, but I feel you're too
old to be at home, we feel you should be away" (p. 178). Baitaxe also found
the autistic subjects violated "conversational postulates" of acceptability and
politeness (Bates, 1976). Their behavior did not suggest they intended to be
rude, but simply that they did not understand the social rules governing what
is acceptable in conversation. A third impairment Baltaxe found was that
the autistic adolescents tended not to "foreground and background" their
utterances. That is, their choice of words did not allow the listener to differen-
tiate between old and new iliformation. For example, they did not use such
devices as definite articles and relative clauses to background old informa-
tion, or use fully specified noun phrases and indefinite articles to introduce
new information. In another study by the same author (Baitaxe & Simmons,
1977) the bedtime soliloquies of an autistic girl were recorded. They found that
the girl tended to make her speech a monologue, whereas normal chil-
dren often act out a two-way conversation (Weir, 1962).
The next study to specifically assess pragmatic abilities in autistic chil-
dren was by Bail (1978). It is unfortunate that this undergraduate disserta-
tion has not appeared in a published form as it contains a number of very
interesting experiments. She found that, compared to matched aphasic chil-
dren, autistic children were more impaired in the range of "speech acts" they
employed (such as relating past experiences, conveying thoughts, commenting
on objects, etc.) and in their understanding of discourse rules (such as
the illegality of nonsequiturs). They were also less likely to use gesture com-
municatively. She concluded that autistic children lack "communicative in-
tent," violate Grice's (1967/1975) Cooperative Principle, and appear not to
understand pragmatic presuppositions.
Langdell (1980) reported that autistic children tend to ask embarrass-
ing questions, such as '~-Iow old are you?" to a stranger in the supermar-
ket, and not recognize that this is not acceptable. Another pragmatic deficit
he noted was the pedantic and formal style of speech frequently heard in
higher-level autistic children, inappropriate to an informal social context.
In addition, autistic children often start to talk to people without first using
boundary markers such as '~Hello" or attempting to engage the listener's at-
tention by trying to establish mutual gaze. He concluded that such examples
reveal autistic children's difficulty in taking another person's point of view.
In an unpublished pilot study, Langdell (1980) found autisti~ children were
impaired in their ability to modify their account of what had happened when
talking to someone who had or had not been present.
Against this picture of communicative deficits in autism it is somewhat
surprising to encounter Needleman, Ritvo, and Freeman's (1980) study, which
found that 24 out of 33 autistic children (73~
municatively. The definition they use of communicative is, however, rather
superficial: "vocally makes requests or unsolicited comments directly to an
individual or responds to questions or comments by more than the minimum
utterance required." (p. 392). It is clear from the studies by Baltaxe (1977)
and Ball (1978) that definitions of considerably more subtlety are required
in order to identify communicative (or pragmatic) competence. The ques-
were using language com-
Social and Pragmatic Deficits in Autism 38"/
tion of how to define the term communicative is of fundamental importance,
and we shall return to it later in the article.
In an interesting single case study, Bernard-Opitz (1982) found that an
autistic child's language did vary as a function of the interlocutor (i.e., whether
it was mother, stranger, or clinician), showing some social sensitivity, while
nevertheless revealing pragmatic deficits, such as perseveration on a topic.
Hurting, Ensrud, and Tomblin (1982) manipulated another variable, name-
ly, listener-response to questions. They found that more conversational break-
downs (discontinuations) occurred if the listener did not ask a question back
to the child, suggesting that the autistic children were unable to maintain
the conversation by themselves. In addition, the autistic children appeared
to use questions as their main device to initiate and continue conversation,
but tended to ask questions to which they already knew the answers. They
appeared not to understand the function of questions as requests for infor-
mation. More generally, a number of authors have noted that autistic chil-
dren do not seem to recognize the function of language which serves to inform
others (Ball, 1978; Caparulo & Cohen, 1977; Cunningham, 1968). On these
grounds, their language has been described as being primarily instrumental
(Cunningham, 1968; Schuler, Fletcher, & Davis-Welsh, 1977).
The difficulty with speaker-hearer roles that Baltaxe (1977) found may
be related to what Fay and Schuler (1980) noted as a difficulty in appropri-
ate turn-taking. For example, autistic children are reported to interrupt a
speaker inappropriately (Pacci-Cooper, Curcio, & Sacharko, 1981, cited in
Layton & Stutts, 1985) and to fail to signal turn-taking using eye contact
(Mirenda et al., 1983). The result is that the autistic person remains either
in the speaker role for too long (Bernard-Opitz, 1982; Paul & Feldman, 1984),
violating (]rice's (1967/1975) Maxim of Quantity, or in the respondent role
for too long (McCaleb & Prizant, 1985). It is also manifested in topics re-
maining "unexpanded" (Fay & Schuler, 1980).
Wetherby and Prutting (1984) analyzed the range of speech acts (Austin,
1962) in autistic children's language. They found autistic children requested
objects and actions more often than normal children did, and protested more.
However, there was a complete absence of speech acts used for requests for
information, for acknowledgments of others, for showing off, and for com-
menting. This supports BaU's (1978) findings. Wetherby and Prutting reported
that the autistic children demonstrated the ability to regulate:an adult's be-
havior to obtain objects, or to obtain an environmental end, but lacked the
ability to attract and direct an adult's attention to him or herself or an ob-
ject as an end in itself.
Tager-Flnsberg's (1981) review article concluded that phonological and
syntactic development in autistic children follows the same course as in nor-
real children Out sometimes at a slower rate), while semantic and pragmatic
functioning appears specially deviant. Her later experiments (Tager-Flusberg,
1985) suggest that representation oi ~ semantic knowledge in autism does not
differ from mental age matched control subjects. Thus the pragmatic deficits
appear to be the main area of deviance in language in autism.
Wetherby (1986) raised the question of whether autistic children are
"noncommunicative." As we remarked in the discussion of Ball's (1978) work,
this naturally begs the question of what is meant by communicative. Wether-
by's (1986) and Prizant's (1983, 1984) definition is using language in an in-
tentional way towards another person to achieve environmental or social ends.
In support of the view that autistic children are communicative, Wetherby
cited Prizant and Duchan (1981) and Prizant and Rydell's (1984) studies
suggest that autistic children's echolalia functions as expressions of inten-
tions to request, protest, affirm, etc. Wetherby and Prutting's (1984) study
also indicated that autistic children use language intentionally (e.g., to ob-
There is no dispute that autistic children can use language intentional-
ly (just as they can use tools intentionally), but is this a full enough defini-
tion to conclude that they are communicative? Speech Act Theory (Austin,
1962; Searle, 1965) defines communication as comprising "complex
intentions"-that is, the speaker's intention to affect the listener's intentions
and beliefs. Under this theory, focusing on the speaker's intentions alone
comprises only half of the definition of communication. We discuss this ques-
tion further in the final part of this paper, when considering a cognitive the-
ory of the social and pragmatic deficits in autism.
It is worth noting that pragmatic impairments are not restricted to
spoken language. Use of gesture is also impaired (Bartak et al., 1975; Cur-
cio, 1978; Attwood, 1984; Wetherby & Prutting, 1984; Ohta, 1987) as is the
comprehension and production of communicative facial expression (Lang-
To summarize, the literature on pragmatic skills in autism presents a
strikingly consistent picture of severely impaired functioning on almost all
aspects that have been tested. This has led to the view from a number of
authors that autistic children use language instrumentally but not communica-
tively. The non-speech-specific nature of the pragmatic impairment supports
the notion that its basis is more than a surface linguistic phenomenon.
In contrast to this picture of pragmatic deficitis, the literature on nor-
mal children's language shows pragmatic competence at a surprisingly early
age. For example, 2-year-olds can adapt their message to what the listener
knows or does not know, and respond to listener feedback (Wellman & Lem-
pers, 1977; Mueller, Bleier, Krakow, Hegedus, & Carnoyer, 1977; Furrow,
1984). Another contrast is with other handicapped groups: Language-delayed
children (Rom & Bliss, 1981; van Kleeck & Frankel, 1981; Ball, 1978) and
Down's syndrome children (Coggins, Carpenter, & Owings, 1983) show the
normal range of speech acts, as do mentally handicapped adolescents (Price-
Williams & Sabsay, 1979; Longhurst, 1974; Bedrosian & Prutting, 1978).
Social and Pragmatic Deficits in Autism 389
Mentally handicapped adults can also comprehend indirect requests, whereas
autistic people cannot (Paul & Cohen, 1985). Nonautistic developmentally
delayed clinical groups thus appear to possess pragmatic competence, in com-
parison to their autistic counterparts.
WHAT MIGHT UNDERLIE
THE SOCIAL AND PRAGMATIC DEFICITS?
The above review makes plain the severity of the impairment in social
and pragmatic skills in autism. Moreover, the research suggests a particular
profile: the impairment affects specific social skills (but not others), and af-
fects almost all pragmatic skills. What underlying psychological mechanism
might be responsible for such a picture? In the final part of this paper we
consider two possible theories that have addressed this question.
The Affective Theory
On a number of levels, autism should be counted a disorder of affective and social
relations-and irreducibly so. (Hobson, in press)
One theory proposes that the social and communication deficits in
autism are primarily affective. This view should not be confused with the
notion that autism is an emotional response to trauma (Bettelheim, 1967;
Tinbergen & Tinbergen, 1983). Rather, the Affective theory states that in
autism there is an innate inability to enter into emotional touch with other
people. This theory was originally proposed by Kanner (1943), as his title
"Autistic Disturbances of Affective Contact" makes clear. The most detailed
version of this view, however, has come from Hobson (1983, 1986a, 1986b,
in press; Weeks & Hobson, 1987), whose work we touched on earlier. We
consider his theory here. Unless otherwise stated, references to him are based
on his most recent work, Hobson (in press).
Hobson summarized his theory in terms of four major axioms. These
(1) that Autistic children lack the constitutional components of action and reaction
as are necessary for the development of reciprocal personal relations with other peo-
ple, relations which involve feelings. (2) Such personal relations are necessary for the
'constitution of an own and common world' with others (Bosch, 1970, p. 115); (3)
Autistic children's lack of participation in intersubjective social experience has two
results which are especially important, namely (a) a relative failure to recognize other
people as people with their own feelings, thoughts, wishes, intentions, and so on,
and (b) a severe impairment in the capacity to abstract and to feel and think sym-
bolically. (4) The greater part of autistic children's cognitive and language disability
may be seen to reflect either lower-order deficits that have a specially intimate rela-
tionship with affeetive and social development, and/or impairments in the social-
dependent capacity to symbolize. (p. 3 original Manuscript).
Lack of innate abiliW to interact
emotionally with others
people's mental states
Impaired ability to abstract
Fig. 1. The affective theory.
This position is expressed diagrammatically as in Figure 1.
Hobson's starting point is that normal infants are prewired to be sensi-
tive to and comprehend another person's emotions. This assumption is drawn
from studies on mother-infant interaction such as that by Murray and Trevar-
then (1985). Their ability to do this, Hobson argued, is "beyond cognition."
Butterworth (1986), referring to Hobson's theory, emphasized this by say-
ing '~he mind is transparent" (p. 20), that is, that other people's mental states
such as their emotions are "naturally" available to us. Biological prewiring
is the solution some philosophers have proposed to the problem of how we
know other people have minds (Hamlyn, 1974). This led Hobson to ar-
gue that other people's mental states do not need to be inferred, but can be
perceived "directly" in their bodily expressions. He calls this "non-inferential
empathy" (p. 12). In autism, Hobson proposed, this biological, noncogni-
tive prewiring for understanding emotional states in others is nonfunctional.
Hobson goes on to propose that the development of a symbolic capac-
ity and of a conceptual role-taking ability are both directly derived from the
infant's affective relationships with others. In such relations, he argues, the
infant comes to appreciate another person's way of conceiving and seeing
an object, and it is this that provides the infant with the notion of symbolic
interpretation and other people's conceptual viewpoints.
How wen does this account make sense of the social and pragmatic
deficits in autism reviewed earlier? Hobson's own studies (1986a, 1986b) which
Social and Pragmatic Deficits in Autism.
investigated autistic children's understanding of emotional expressions are
some support for his theory, although his tasks required intermodal recog-
nition of emotions (gestures, vocalizations, contexts, and facial expressions),
and it is unclear which component in all this may have caused their failure.
Langdeil's (1981) results suggested that when only one modality (facial ex-
pressions) is tested, their performance is still impaired, but is above chance.
Similarly, 17 of the 23 autistic children in Hobson's (1986a) experiment could
match schematic facial expressions with videotaped facial expressions, after
some teaching, and Hobson (1986b) found they could match drawings of
gestures with films of gestures. This suggests that the difficulty was in match-
ing facial, vocal, and gestural emotional expressions. Weeks and Hobson's
(19~7) result suggested autistic children can recognize facial expressions but
do not do so preferentially.
Axiom 3(a) of Hobson's model predicts that conceptual role-taking
should be impaired in autism, as has been found (Baron-Cohen et al., 1985,
1986). However, it is not clear why his model should make this prediction,
as difficulty in understanding emotions does not necessarily imply difficulty
in understanding beliefs. Nor does his model account for unimpaired func-
tioning in perceptual role-taking (Baron-Cohen, 1985; Hobson, 1984) or in
self-recognition (Flannery, 1976; Neuman & Hill, 1978; Ferrari & Mathews,
1983; Spiker & Ricks, 1984; Dawson & McKissick, 1984; Baron-Cohen, 1985).
In addition the Affective theory does not account for why attachment in au-
tism may to some degree be unaffected (Sigman & Ungerer, 1984), or why
the social smile at 6 weeks of age may be present in autism (Park, 1983) nor
why autistic children may enjoy rough-and-tumble play (Damasio & Maur-
er, 1978). 3
Asiom 3(b) suggests that one direct consequence of a nonfunctional
innate ability to perceive people's emotional states is an inability to abstract
and symbolize. 4 Hobson used this claim to account for autistic children's
deficits in pretend play (Baron-Cohen, 1987a; Ungerer & Sigman, 1981), in
3Unpublished data we have collected from parental reports of 22 autistic children in Britain
reveals that 81070 smiled at 6 weeks of age, and 50% played "peek-a-boo" games, enjoyed
cuddling in infancy, and became upset when a parent left. Volkmar (1987) also reported
that 50~ of their sample (N = 50) of autistic children smiled socially in infancy, and enjoyed
cuddling. Delong (1978) reported that only 3 of 17 autistic children (18070) had a social smile,
as parents recalled. Clearly, more reliable methods of assessing these early social skills are
needed, but such findings, if confirmed, pose a problem for the Affective theory.
4Hobson (personal communication), commenting on Fig. 1, suggested the arrow from the Wailure
to recognize other people's mental states" box to the UImpaired ability to abstract and symbolize"
box, as he sees the latter as the developmental oUtcome Of il/e former as well as of the
top box. I am grateful to him for this improvement in the accuracy of the figure.
abstract imitation (Curcio, 1978; Hammes & Langdell, 1981), and in lan-
guage. However, the mechanism by which the development of a symbolic
capacity necessarily "depends upon the infant's experience of a world of
shared feelings and patterns of activity with others" (p. 14) and is "essential-
ly affective-conative and/or social in origin" (p. 20) requires much more
clarification and empirical evidence than is presently available.
The Meta-Representation Theory
In contrast to the Affective theory, we have proposed a primarily cog-
nitive explanation for the social impairment in autism (Baron-Cohen et al.,
1985). This is not the only cognitive theory of autism that has been formu-
lated (e.g., see Hermelin & O'Connor, 1970; Boucher, 1981; Rutter, 1983),
but it is referred to as the Cognitive theory in order to distinguish it from
the Affective theory. We discuss its assumptions here; for purposes of quick
comparison with the Affective theory we then summarize the Cognitive the-
ory as five axioms and as a diagram (Fig. 2).
Like the previous theory, the Cognitive theory also considers as cen-
tral the autistic child's difficulty in understanding other people's mental states.
However, unlike the Affective theory, this view starts from the premise that
mental states are not directly observable but have to be inferred, an infer-
ence that requires a complex cognitive mechanism which is described later.
The Cognitive theory also places more emphasis on the ability to infer men-
tal states such as beliefs, rather than emotions, for the following reasons:
Beliefs and desires are held to be the most important mental states in mak-
ing sense of the social world, because they have a causal relationship to ac-
tion (Dennett, 1978), and they have this by virtue of their content; beliefs
and desires are always about something (i.e., I believe that x, and I believe
you believe that y). This "aboutness" of mental states is termed its Inten-
tionality (Brentano, 1874). Unlike such mental states as beliefs and desires,
emotional states (such as happiness, sadness, fear, anger) do not necessarily
have content, and as such may be of less use in predicting and making sense
of social behavior.
The ability to attribute mental states with content to others has been
called a '~heory of mind" (Premack & Woodruff, 1978), because it involves
the person postulating the existence of mental states and then using these
to explain and predict another person's behavior. Dennett (1978) and others
argued that we use such a folk psychology all the time to make sense of the
social world (e.g., "He won't talk to me because he believes I don't like him,"
Wimmer and Pemer (1983) and Hogrefe, Wimmer, and Perner (1986)
found that normal children of 4 years of age can attribute a false (and there-
Social and Pragmatic Deficits in Autism
fore different) belief to another person (e.g., the chocolate is in the cupboard)
and can use this to predict the person's behavior (she will look in the cup-
board), as can Down's syndrome children (Baron-Cohen et al., 1985). In con-
trast, autistic children of normal intelligence failed to demonstrate that they
could distinguish their own belief from someone else's (Baron-Cohen et al.,
1985; 1986). This is seen as an autism-specific deficit, and has been confirmed
by others (Dawson & Fernald, 1987).
As mentioned earlier, the inferential operation involved in attributing
mental states such as beliefs is held to require complex cognitive structures.
Their basis can be summarized as follows: Our beliefs about or concepts of
the physical world may be called "primary representations." However, our
beliefs about other people's mental states (such as their beliefs and desires)
are representations of other representations. These may be called "second-
order representations" (Dennett, 1978; Johnson-Laird, 1983) or "meta-
representations" (Leslie, 1987; Pylyshyn, 1978). 5 Primary and metarepresen-
tations have very different logical properties (Leslie, 1978). The Cognitive
theory posits that in autism the capacity for meta-representation is impaired.
Leslie (19_87) outlines the way in which such meta-representations
may operate to allow not only attribution of different beliefs and desires to
another person but also pretend play. In the latter, the cognitive system must
simultaneously represent an object as real and unreal. Pretend play is not
immediately relevant to this article but is discussed more fully elsewhere
At what age would one normally expect a capacity for meta-
representation to develop? Neither the ability to attribute different
beliefs to another person nor the ability to pretend play (or symbolize) have
been convincingly shown to be within the repertoire of infants until at least
the end of the first year of life. Current evidence suggests that in normal
children the ability to pretend precedes the ability to attribute beliefs to others
(for a review of this, see Leslie, 1987).
What sense does this theory make of the literature on social deficits
in autism? The theory predicts that only those social skills requiring a meta-
representational capacity should be impaired, such as conceptual role-taking.
Our earlier studies support this prediction (Baron-Cohen et. al., 1985, 1986).
In addition, the lack of any self-conscious reaction in the mirror self-
recognition studies in autism is explained in terms of an inability to con-
ceive of oneself as the object of another person's thoughts. Indeed, this ex-
planation can be extended to the lack of embarrassment in autism ingeneral
(Baron-Cohen, 1985). Mundy et al. (1986) used the theory-of-mind explana-
5Elsewhexe we have used the term second-order representation (Baron-Cohen, 1987a). Here we use
its synonym meta-representation because it is marginally less cumbersome, and is consistent
with other authors (Leslie, 1987).