Book Reviews 
Natural Language Processing 
Harry Tennant 
Petrocelli Books, Princeton, 1981, 
276 pp., Paperback, $17.50, ISBN 089433-100-0. 
(Dist: Van Nostrand-Reinhold, New York) 
How do computers understand natural language? 
This question is asked by interested lay persons as well 
as computer scientists who are not familiar with artifi- 
cial intelligence. This book provides the reader with 
an introduction to the field and the general flavor of 
many natural language processing programs. The em- 
phasis is on the positive results that have been attained 
and general methods that are frequently employed. 
The book opens with motivations for natural lan- 
guage research: many practical advantages for user 
interfaces and understanding the human mind. Subse- 
quent chapters deal mainly with the first motivation. 
A description of a simple natural language processor is 
the vehicle used to introduce the concept of semantic 
nets, rudimentary syntactic and semantic analysis, and 
the inherently ambiguous nature of natural language. 
Chapter 2 surveys a variety of early natural language 
processing programs, including BASEBALL, SAD- 
SAM, SYNTHEX, STUDENT, DEACON, and DOC- 
TOR. Unfortunately, no rigorous analysis of the limi- 
tations of these programs is presented, other than not- 
ing the limited computing technology available at the 
time and the necessary failure of a word-by-word 
translation approach. 
Chapter 3 is devoted to syntax, using BNF nota- 
tion, transition networks, surface structure, deep struc- 
ture and transformations (though no mention of 
Chomsky!), followed by discussions of top-down, 
bottom-up and hybrid parsing schemes. Chapters 4 
and 5 present semantics. Chapter 4 is an excellent 
introduction to the types of problems a semantic com- 
ponent of a natural language processor must address. 
For the person who has not thought deeply about the 
nature of natural language, the examples clearly distin- 
guish problems such as multiple word senses, modifier 
attachment, noun-noun modification, pronouns, deter- 
miners, ellipses, and substitution. Case frames and 
concept decomposition are presented as means for 
dealing with these problems. Chapter 5 is a large 
collection of brief descriptions of implemented seman- 
tic analyzers: Air Line Guide, ROBOT, Preference 
Semantics, SOPHIE, LIFER, Linguistic String Project, 
PLANES, RENDEZVOUS, and ELI. Each semantic 
analyzer is presented independently; no analysis of 
strengths and weaknesses is given. 
The notion of frames is presented in Chapter 5, 
along with a good discussion of understanding as a 
memory-based process and the close coupling of natu- 
ral language processing and knowledge representation. 
Throughout the first six chapters, natural language 
understanding is presented as single sentence under- 
standing. The final chapter, however, addresses dis- 
course analysis: speech acts, rules for discourse and 
the difference between the rules for spoken and writ- 
ten dialog. 
I would recommend this book to the general reader 
as a comprehensive introduction to natural language 
processing that does not get bogged down with details 
and the inner workings of particular implementations. 
The reader will come away with an understanding of 
both the enormous problems natural language under- 
standing researchers must address and the many gener- 
ally accepted methods of attacking these problems. 
The reader will not, however, be provided with under- 
lying linguistic theory, a unifying historical perspective 
or an analysis of inadequacies. 
Although the book is flawed by a lack of theory 
and the use of formalisms, such as semantic nets and 
BNF, without definition, it might be a good text for a 
stand-alone introductory undergraduate AI course, if it 
is supplemented with further readings and lectures that 
provide more depth. 
Sharon C. Salveter, SUNY Stony Brook 
