Book Reviews The Process of Question Answering - A Computer Simulation of Cognition 
The Process of Question Answering - 
A Computer Simulation of Cognition 
Wendy G. Lehnert 
Lawrence Erlbaum Associates, Hillsdale, N.J., 1978, 
278 pp., $16.50, ISBN 0-470-26485-3. 
This book can be reviewed from two perspectives: 
(1) as a general and technical introduction to the com- 
plexities and subtleties regarding human understanding 
of and response to questions and (2) as a much more 
comprehensive system, intended to produce not only 
humanlike interpretations of questions, but doing so in 
ways that model human thought processes. Lehnert 
appears to have both approaches in mind, emphasizing 
the second more, but it is with respect to only the first 
that the book is successful. 
From this first perspective, then, there is no other 
single source that so completely and persuasively illus- 
trates the many-faceted problems of translating a 
person's question (e.g., "Do you have a match?" vs. 
"Do you have a hangover?") into a valid interpreta- 
tion of the questioner's desires or information require- 
ments (e.g., "Give me a light" vs. "How is your physi- 
cal state after last night?"). Lehnert provides abun- 
dant evidence for the necessity of utilizing a wide 
American Journal of Computational Linguistics, Volume 6, Number 3-4, July-December 1980 187 
Book Reviews The Process of Question Answering - A Computer Simulation of Cognition 
variety of information in the analysis of questions. 
This information can lead to the transformation of the 
immediate naive understanding of the question into a 
drastically changed conceptualization. 
There are four major components to the question- 
analysis and answer-generation approach she de- 
scribes: (1) different kinds of knowledge sources, (2) 
a variety of mapping and inferencing principles, (3) a 
parser which converts the surface text into one of a 
small number of conceptual representations (actually, 
Schank's system), and (4) a structured taxonomy of 
question types, one of which is assigned to the parsed 
representation. Among the most important knowledge 
sources are those involving "scripts" and contexts. 
The well established notion of "scripts" refers to the 
fact that many human activities have a stereotypical 
pattern of actions and events (e.g., eating at a restau- 
rant: "look for table, go to table, ask for menu, get 
menu, be served," from Fig. 6.1, p. 141). Given that 
a script is judged to apply (e.g., cued by a verbal nar- 
rative beginning "John went into a restaurant"), Leh- 
nert indicates how the details of the script can provide 
the basis for interpreting and responding to a wide 
variety of questions (e.g., "What did John do next?", 
"Did John eat?"). The importance of prior communi- 
cation and the physical setting are also stressed as 
contextual sources of information for establishing top- 
ics or references that would otherwise be indetermi- 
nate. Two other knowledge sources, stressed as im- 
portant but worked out in less detail, are general 
world-knowledge and personal memories. 
Lehnert outlines as the second major system com- 
ponent a number of specific infereneing mechanisms 
for hypothesizing an initial question-type, for utilizing 
knowledge sources to translate this into a more valid 
representation, and for generating the answer. Indeed, 
she invokes, in one way or another, most of the impor- 
tant concepts involved in artificial intelligence. The 
use of inferencing principles and alternative knowledge 
sources can be governed by rather flexible search heu- 
ristics, but the power and accuracy of her approach 
depend critically on the use of a small number of 
conceptual-representation and question categories, the 
third and fourth system components. While a number 
of technical questions can be raised as to the adequacy 
of these categories, she does provide many examples in 
support of these, illustrating how fine distinctions 
among conceptualizations and question-types might be 
made. 
Were Lehnert's goals only those described as the 
first review perspective, she would have achieved them 
rather remarkably: Many thorny problems in question- 
answering are presented, and she illustrates in some 
detail the processing algorithms that could be applied 
to provide a solution; she also indicates numerous 
areas where further work is needed. There is no ques- 
tion but that Lehnert provides an important and de- 
tailed introduction to these problems as well as a sig- 
nificant challenge to both psychologists and computer 
scientists for testing or extending the theoretical ideas 
proposed. 
Lehnert's goals were, however, much more ambi- 
tious, as indicated by the book's subtitle, and through- 
out. In the preface, having shortened the phrase 
"human information processing" to "information proc- 
essing," she states that "this book describes question 
answering as a particular task in information process- 
ing" (p. viii). In the first paragraph, she says, "This 
thesis presents a process model of question answering 
as a theory of conceptual information processing" 
(p. 1). In her final sentence, she summarizes by saying 
that her system is "a theory of question answering that 
is founded on and extends theories of natural language 
and conceptual information processing" (p. 270). 
On the basis of these very strong claims, the reader 
may justifiably expect to find supporting evidence of 
three kinds: (1) that the syntactic and semantic prob- 
lems of question-interpretation identified and dis- 
cussed extensively by linguists will be acknowledged 
and dealt with, (2) that existing psychological evidence 
on language and problem-solving behavior in general, 
and question-answering in particular, will be reviewed 
and related to the processing principles of the system, 
and (3) that performance statistics will be provided for 
a wide variety of test questions, with comparisons 
made of the system's interpretations and intermediate 
processing steps to those of humans. 
In fact, none of these expectations is fulfilled. 
With respect to linguistic data, there is almost no ref- 
erence at all to the enormous body of literature that 
exists on the interpretation of interrogatives (e.g., 
Aqvist's A New Approach to the Logical Theory of In- 
terrogatives, 1975, and Hudson's The Meaning of Ques- 
tions, 1975). Even without discussion of the particular 
views of linguists, there still are well-known technical 
linguistic problems which surely should be addressed 
by systems intended to implement computer processing 
of questions, such as whether noun phrases are to be 
understood referentially or attributively, determining 
qualifier scope, and interpreting quantification (see the 
Syntax and Semantics book series by Academic Press). 
The notable exception is her criticism, in a discussion 
of semantics, of Katz and Fodor's (1964) theory as 
being inadequate for semantically representing the 
equivalence of active-passive transformations (p. 249). 
She fails to report, however, that Katz's later book 
(Semantic Theory, 1972), which is an extension and 
modification of his earlier work, does in fact provide 
for exactly the type of representation for which she 
seemed to be arguing. 
With respect to psychological data, no reference is 
made to the sizable body of literature concerning the 
development of human question-asking (e.g., Davis, 
188 American Journal of Computational Linguistics, Volume 6, Number 3-4, July-December 1980 
Book Reviews The Process of Question Answering - A Computer Simulation of Cognition 
"The form and function of children's questions," 
1932), nor to work concerning adult question- 
answering (e.g., Wright, "Some observations on how 
people answer questions about sentences," 1972), nor 
even to work closely paralleling her own approach in 
the use of a question taxonomy (e.g., Kearsley, 
"Question and question asking in verbal discourse: A 
cross-disciplinary review," 1976). Even more distress- 
ing is the fact that the human problem-solving litera- 
ture, larger by orders of magnitude than the specific 
work on question-answering, is not reviewed to pro- 
vide a comparison for her system's detailed search and 
inferencing processes. 
Finally, almost no performance data of any kind 
are provided, not even the complete details of the 
processing for a single example. In view of the incom- 
pleteness of presentation with respect to the exact 
details, one has no way of ascertaining the limitations 
of her system. Since most of the examples given were 
trivial with respect to the syntactic complexity of ques- 
tions for which other question-answering systems have 
been designed, and since the system appears to be 
very modest with respect to the size of its knowledge 
structures relative to more representative situations, 
there still remains considerable question whether or 
not her system is designed on sufficiently extendible 
principles to permit handling of these more realistic 
instances. 
Despite Lehnert's failure to provide support for her 
more extravagant claims, the reader who adopts the 
less demanding perspective will be well rewarded. 
Lance A. Miller, IBM Research 
\[Reprinted by permission from Contemporary Psycholo- 
gy, 24, 10, 1979, pp. 777-779. Copyright 1979 by the 
American Psychological Association.\] 
The Process of Question Answering - 
A Computer Simulation of Cognition 
Wendy G. Lehnert 
Lawrence Erlbaum Associates, Hillsdale, N.J., 1978, 
278 pp., $16.50, ISBN 0-470-26485-3. 
This attractive and well-written book describes 
QUALM, a facility for question-answering developed 
by the author in conjunction with systems designed by 
Roger Schank and others at Yale. But it is a good 
deal more than a system description, and it will be 
read with interest and profit by logicians (particularly 
for its contributions to the logic of questions and an- 
swers), students of mind and of the philosophy of 
science (for its defense of the methods used in artifi- 
cial intelligence research), and, in fact, all the readers 
of this polyglot newsletter \[i.e. CBT\]. Someone teach- 
ing a course that touched on AI techniques might use 
this as a concrete example, suitable for intensive study 
-- the writing is clear, full of intriguing examples, con- 
cerned with serious problems, and accessible to read- 
ers without any background at all in computer science. 
The logic of questions and answers has developed 
only recently. (For an excellent introduction to this 
"erotetic logic," with an extensive bibliography, see 
Nuel D. Belnap, Jr. and Thomas B. Steel, Jr., The Log- 
ic of Questions and Answers, New Haven: Yale Univer- 
sity Press, 1976.) Inspired by the need for realistic 
dialogue with users of a natural language system, it has 
attempted to identify the presuppositions that enter 
into questions and appropriate responses to them. To 
take one of the hoariest examples, "Is the present king 
of France bald?" presupposes that there is now a king 
of France (Belnap and Steel, p. 110). If we are going 
to build a system that responds helpfully to this ques- 
tion, it is going to have to have some way of identify- 
ing the presupposition. 
But this is only one of the questions that Lehnert 
addresses, and not the most interesting or important 
one. The role of context, the appropriateness of the 
answer, the focus of the question, the memory repre- 
sentations that are required, the inferences that the 
answerer must make, search strategies -- these and 
many other problems are raised at the beginning of the 
book. The remaining chapters describe the way in 
which she solves them. 
QUALM reflects the concepts and approaches that 
Schank and others have developed at Yale. (Cf. R.C. 
Schank and R.P. Abelson, Scripts, Plans, Goals and 
Understanding, Hillsdale, N.J.: Lawrence Erlbaum 
Associates, 1977.) These notions, including the con- 
ceptual dependency approach, and the use of scripts to 
represent expected behavior in a situation, are de- 
scribed in sufficient detail to permit the reader to fol- 
low Lehnert's discussion without previous exposure to 
them. 
I found that the book helped greatly to clarify the 
current status of natural language processing, docu- 
menting the tremendous strides that were taken during 
the 1970's. (The contrast between this book and Ter- 
ry Winograd's pioneering Understanding Natural Lan- 
guage, New York: Academic Press, 1972, is quite 
overwhelming.) Equally interesting, I think, is the 
degree to which Lehnert gives us a sense of what AI 
research is doing, why it thinks that it is a science, and 
what counts as a successful experiment in AI. 
John M. Morris, Clinton, New York 
\[Reprinted from Cognition and Brain Theory, Ill, 3, 
Spring 1980, pp. 113-114, by permission of the author 
and the Editors.\] 
