Book Reviews Meaning and Discourse - A Computer Model of Psychoanalytic Speech and Cognition 
Meaning and Discourse - 
A Computer Model of Psychoanalytic 
Speech and Cognition 
John Henry Clippinger, Jr. 
The Johns Hopkins Univ. Press, Baltimore, 1977, 
231 pp., $15.00, ISBN 0-8018-1943-1. 
Clippinger's book on discourse is aimed at linguists, 
psychologists, psychiatrists, and researchers in artificial 
intelligence, and in 100,000 words, almost totally de- 
void of humor, he takes on Freud, Piaget, Bateson, 
Colby, and Schank; he offers homage to HACKER 
and kudos to CONNIVER; he ignores both linguistics 
and AI work in natural-language generation; he in- 
vents a grammar of English; he performs validation 
tests on a hand-simulated program; and he closes by 
warning us about ignoring the social impact of com- 
puters in the future. All of this is background to a 
program that models one, halting paragraph of speech 
by a depressed patient whose request to change the 
form in which she pays her therapist is, we are told in 
great detail, a desire for intercourse. 
With regard to the process of producing language, 
much of what Clippinger says is quite true, even en- 
couraging. Real generation is not a simple mapping of 
meaning representations into surface strings. Rather, 
there is an enormous influence by other contextual 
elements such as the speaker's intentions and the 
speaker's model of the hearer's possible reactions, as 
well as the problems of lexical choice and ordering, 
compounded by the incessant demand to say something 
despite a storm of conflicting, unorganized interests. 
Here, Clippinger's high-level descriptions are useful in 
reminding -- or informing -- us about the complica- 
tions in actual speech, which are all the more obvious 
in a psychiatric patient. He specifies a 4-point 
"theory of thinking and speaking": (1) knowledge 
structures are organized by goals; (2) beliefs can be 
inconsistent; (3) meaning depends on the context of 
interpretation; (4) composing and producing a dis- 
course occurs in 5 steps: identification of an accepta- 
ble goal, planning the actual discourse, considering the 
reaction by the listener, choosing the exact words, and 
reviewing the final product. This, of course, Works for 
writing, too, or for that matter, for general problem- 
solving, where carrying out physical acts substitutes 
for choosing words. As an illustration of the complex- 
ity of cognition, this text serves well. 
The weaknesses of the book are in the lower-level 
detail, and it is disappointing not to see the theory 
better illustrated, especially since these details form 
the bulk of the book. 
The representation of meaning is hardly different 
from the surface form. "George painted the house" is 
represented as something on the order of (GEORGE 
PAINT HOUSE). This can be modified with any of a 
large number of "indicators" including the customary 
ones like time and place, but also some problematic 
ones like "condition," which is defined as "those con- 
cepts that must be valid in order for the indicated 
concept to be valid," a can of worms if ever there was 
one. What would one do with the sentence "I must 
leave at noon"? Is "at noon" a time or a condition? 
The concepts are manipulated by pieces of the pro- 
gram, whimsically labeled Leibnitz, Freud, Calvin, 
Machiavelli, and Cicero. Calvin censors potential 
output, Machiavelli plans, Cicero orates, and so forth. 
Chapter 5 describes the "grammar for discourse" as 
a series of rewriting rules. Clippinger's protests not- 
withstanding, the grammar describes little more than 
surface syntax. With examples like "It seems like only 
yesterday, I left my mind behind" anyone would be on 
dangerous semantic grounds. 
The program that "generated" a text resembling 
the actual speech by the psychiatric patient is barely 
sketched out in Chapter 6. The program implements 
only a small part of the theory, which is no surprise, 
but Clippinger's apology is disingenuous: 
The purpose in designing and writing the ERMA 
program was to test out a theory of the discourse 
formation and realization processes. In writing the 
program, the entire discourse episode was analyzed 
and hand-simulated. However, only part of the 
entire discourse episode was computer-simulated. 
The reason for this is simply that since all the ma- 
jor representational and theoretical problems had 
been worked out, the additional programming need- 
ed to make the program complete would not have 
contributed substantively to either the theory or the 
methods developed. It is my belief that, as a re- 
search strategy, little is gained, and often much is 
lost, by developing massive operational programs 
when the research is still in an exploratory stage. 
\[page 145\] 
In the last chapters, Clippinger expresses the view 
that computer models of thinking may help psychiatry 
by contributing a vocabulary, a new, more precise way 
of thinking, and testable theoretical models, which 
may indeed be the case some day, a phenomenon not 
unknown to computer science. But he gets quite car- 
ried away with an analogy between aspects of our 
mental processes and the context mechanism in CON- 
NIVER, described in terms that would make McDer- 
mott and Sussman blush. Arguments along the line of 
"My program fumbles and so do people so my pro- 
gram is a model of thinking" are hopelessly reduction- 
ist. 
In all, this is a very ambitious work, and Clippinger 
has much to say on a great many topics, so it is per- 
haps less surprising that there is much to disagree 
with. But all readers should benefit from the elabo- 
rate reminder of the complexities of human speech. 
James R. Meehan, University of California, Irvine 
American Journal of Computational Linguistics, Volume 6, Number 2, April-June 1980 113 
