Computer B
GARY
Content J.
TUCKER.
Analysis M.D.,
AND
of Schizophrenic
STANLEY
D.
ROSENBERG,
Computer analysis signtficantly dtfferentiated the thematic content ofthefree speech oflO schizophrenic patientsfrom that oflO nonschizophrenicpatienzs and f rom the content oftranscripts ofdream material from 10 normal subjects. Schizophrenic patients used the thematic categories infactor I (the “schizophrenic factor”) 3 times morefrequently than the nonschizophrenics and 10 times morefrequently than the normal subjects (p < .01 ). In general, the language content ofthe schizophrenic patient mirrored an almost agitated attempt to locate oneselfin time and space and to defend against internal discomfort and confusion. The authors discuss the implications ofthis studyforfuture research.
Speech:
A Preliminary
Report
PH.D.
burden of attempting to establish reliability without sacrificing meaning (2, pp. 469-503; 5; 7- 10). Most quantitative analyses have focused on material written by patients (which may be very different from free speech), and most have used scoring techniques that require many subjective judgments. However, whether they focus on content or formal structure, all of these studies ask the same question: What makes this speech “crazy,” unique,” or “different”? Theme must be something in the free speech of these patients that we can characterize. Brown ( I I ) recently noted that even the most bizarre or delusional content can be made credible and lose its “crazy label” if we preface it with the statement, I dreamt.” At the same time, it seems that the spontaneous verbal productions of the prospective patient are a primary cue that clinicians use in their decision to affix the diagnosis of schizophnenia to a given case. Moreover, this diagnosis is a fairly common one. There is something (or a combination of things) that a person can say to make the clinician infer that he suffers from schizophrenia. We are concerned with specifying the linguistic cues that the patient presents to trigger this inference. In our study we have tried to combine modern computer methods and analysis with a phenomenologic study of the experience of schizophrenic patients as expressed in their free speech. We have attempted to answer the following questions: I) Can the content of schizophrenic speech be differentiated from that of nonschizophnenic speech as well as from the content of dream material? 2) Can a reliable, objective, and simple technique to do this be devised through the use of modern computer technology? “
“
As THE TREATMENTS for psychiatric illnesses have become more specific, we have seen the interest of psychiatnists return increasingly to the phenomenology of mental illness. Regardless of the patient’s own history, family history, or even the presenting clinical picture, we place a great deal of emphasis (as Rosenhan’s widely publicized recent study has shown [1]) on precisely what the patient tells us or complains of. However, there is probably no phenomenological area more difficult to study in a systematic way than the mass of verbal productions that a patient presents to us. There have been many attempts to study the language of patients, particularly of schizophrenics. Most of these studies have focused on two areas: the evaluation of the formal or structural aspects of language (grammar, syntax, etc.) (2, pp. 469-503; 3; 4) and the examination of content or thematic differences (5-10). The latter is the research technique most closely akin to the clinical interview and to everyday practice. While clinical experience argues that an understanding of language characteristics is crucial to a phenomenology of psychosis, a number of difficulties have limited progress in this area. Quantitative studies of the form or content of language have traditionally required many laborious hours of scoring and have struggled under the
Revised American
version of a paper presented Psychiatric Association,
Dr. Tucker is Professor partment of Psychiatry, 03755.
at the Detroit,
127th Mich..
annual May
and Dr. Rosenberg is Assistant Dartmouth Medical School,
This work was supported by General Research 05392 and S9l-l44 from the General Research sion of Research Facilities and Resources, Health.
Support Support National
meeting of the 6-10, 1974. Professor, Hanover, grants Branch, Institutes
DeN.H. RRDiviof
METHOL)
Subjects
and Setting
The subjects included in this study consisted of the following three samples: 10 schizophrenic patients, 10 nonschizophrenic patients, and 10 normal subjects. The technique used in the selection of these subjects and the interview method have been previously reported (8-10). The only criteria for the selection of subjects focused on making the three groups comparable in age, sex, and social class. The normal subjects were selected on the basis
of
1) exhibiting
no
overt
pathology
and
2)
demo-
graphic similarity with the two patient groups. The schizophrenic and nonschizophrenic patients were sequential admissions to the acute inpatient service of a general hospital. The diagnosis of these patients was arrived at by
AmJ
Psychiatry
132.6,June
1975
611
ANALYSIS
OF
SCHIZOPHRENIC
SPEECH
consensus of two clinicians, who classified the patients according to APA diagnostic criteria (12) as well as by the New Haven Schizophrenia Index (I 3). The diagnoses of the nonschizophrenic patients included neurotic depression, character disorder, etc. All of the patients could be classified as acutely ill, and the schizophrenic patients would be classified as reactiverather than process-type schizophrenics.
Each
of
the
three
groups
consisted
of
5
men and 5 women. The median age of the three groups was 20 years, and the subjects fell into the range of lower to upper middle-class. The transcripts of dream material from the normal subjects were obtained as part of a larger study of the normal dream activity of individuals who were panticipating in sleep studies.’ The subjects were awakened after they had experienced 10 consecutive minutes of rapideye-movement sleep and were asked promptly to describe their dreams. The transcripts from the patients were randomly selected from a larger sample of transcripts previously collected (8-10); they were obtained during a standardized 15-minute interview during the patients’ first week of hospitalization. All of the patients were given the same instructions; after this introduction the interviewer, an experienced psychiatrist, spoke only when he thought it necessary to encourage the patient to continue talking, and his comments were limited to that goal. This technique has been previously described and used in other studies (8-10). The tapes were then transcribed to typescnipts
tent
and
connected
by
the
interviewer
prior
from
A nal6sis
alphabetical words,
allow
available Dictionary
remove these
suffixes words
such
to
then
dictionaries. allows 3,500
The
as be
“-ed”
searched
Harvard
III
words (most of the words used in everyday speech) to be placed into 84 thematic categories that have been specifically selected for their psychological and sociological relevance (see ap‘These transcripts were provided by Dr. Peter Psychiatry, Dartmouth Medical School.
612
vsis
The output of these computer programs can be listed as the combined frequency of the appearance of words in each of the 84 thematic categories. The program also has the capability of identifying specific sentences in any of the typescnipts and also the appearance of specific thematic categories in any sentence in the typescripts. Consequently, individual themes and sentences can be exammed in their actual context. The combined frequencies of the three groups of subjects were then studied by a oneway analysis of variance and a factor analysis. In this particular study we were concerned with each subject’s range of vocabulary rather than his possible repetition of any given word. The issue of repetition of words in schizophrenic speech has been previously studied(15). Consequently, our analysis was based on the subjects’ initial use of each word but disregarded the subsequent use of the same word. Later studies will examine the relationship between the range of vocabulary and repetition of words.
Thematic
order, and
for in various Psychosocial
Anal
RESULTS
These samples of free speech were analyzed through the use of a Dartmouth adaptation of the General Inquiren Computer Content Analysis Program and the Harvard III Psychosocial Dictionary (2, 14). Although many of the transcripts were fairly long, only 600 message units or about 600 words were placed into the computer for each subject for analysis. This limitation was imposed by the storage capacity available to us. To ensure comparability and prevent selection bias, the first 600 words ofeach transcript were used, and this generally represented a large percentage (usually more than twothirds) of the entire interview. The General Inquirer Computer Content Analysis Program simply allows these typescnipts to be processed by the computer so that they can be grouped into various thematic categories depending on the dictionary that is used. The General Inquirer program is actually a series of programs that does such things as arrange the words into
Data
to con-
analysis.
Content
pendix I). They cover such areas as emotional states, thoughts, evaluation, and cultural objects and settings. A fuller description of the programs and the dictionary can be found in The General lnquirer(2). Using this technique we were able to “tag,” or categorize, over 90 percent of the words used in these typescripts.
Am J Psychiatry
132.6,
June
Hauni,
1975
Department
of
Categories
Seventeen thematic categories significantly differentiated among the speech samples of the three groups (table 1 ). Fourteen categories significantly differentiated schizoph renic patients from nonschizoph renics. These categories can be grouped into the following three general areas: those that represent I) heightened emotional states, such as the categories of avoidance, attack, overstatements, and sense; 2) orientation to people in the environment, with categories denoting such things as references to space, time, and sensation; and 3) attempts to explain their condition, with categories such as get, cause, ought, and self. The nonschizophrenic patients were significantly diffenent in their more frequent use of the categories of ideal, if, and medical. It is interesting that the dream subjects scored significantly higher on only the category of urge. This category includes such terms as drive states, dream, eager, and incentive. Ifwe now look at the magnitude of mean differences among the groups, we see that in 10 categories the dream subjects and nonschizophrenic patients were more similar to each other than they were to the schizophrenics (see table 1 ). In many ways, then, it appears that speech content associated with schizophrenia is less like the content of dream material than we might assume. The dream subjects used 5 categories much less often than the two patient groups (self-reference, ideal-value, if, cause, and medical); the nonschizophrenic patients’ use of 2 of these
GARY
TABLE I Frequency of Three Groups
Category
the
Nonschizophrenics
Self-reference Neuter role** Time reference** Space reference** Ideal-value Urge Sense** If Equal Cause Ought** Attack** Avoid** Attempt** Get** Medical Overstate** By analysis *significantly * *
and dream
(‘ategorie.s
2.80 I .40 10.30 7.40 3.00 0.50 2.00 3.50 0.60 1.70 0.50 0.40 1.00 0.50 3.70 1.70 7.50 of variance. differentiated
SignificantI’
Differentiating
A mong
Dream Subjects
Schizophrenics
I .60 I .20
3.80 2.50
7.90
17.50
pez.Ol
7.50 1.60 1.40 2.80 1.80 0.10 0.90
1 1.00 2.90 1.00 4.10 3.20 1.00 1.90 1.80 2.30 2.20 1 .40 5.00 1.10 10.90
p< .08 p