Inside computer understanding: Five programs plus miniatures Ed. by Roger C. Schank and Christopher K. Riesbeck (review)
Janine Scancarelli · Language · 1982
494 LANGUAGE, VOLUME 58, NUMBER 2 (1982) under random successive drawings?' (p. 40). In no case is the student given the opportunity to create numbers from raw data, or to manipulate data in order to test linguistically significant hypotheses. Only a very few of A's examples concern the social side of language—an oversight which greatly reduces the usefulness ofhis book, since students of sociolinguistics surely form a large proportion ofthe potential clientele fora statistics course within a linguistics program. A minimal prerequisite to the second part of the text (random variables, mathematical expectation , multivariate distributions) is a knowledge of calculus and an ability to interrelate abstract mathematical concepts. Examples and exercises are not stated in linguistic terms, leaving the student with no motivation whatever from the viewpoint of the field for which the text is intended. It is doubtful that linguists will find this part ofthe book relevant—even though it presents some potentially useful concepts, e.g. covariance and correlation. For any property that can be counted or measured , anappropriate statisticaltechniqueexists. A's volume, however, seems to have inverted priorities, counting and measuring linguistic externals simply because they happen to be amenable to well-established techniques. A thorough and well-motivated introduction to statistics for linguists is still lacking. [Anthony J. Naro, Universidade Federal and Pontificia Universidade Católica, Rio de Janeiro.] Inside computer understanding: Five programs plus miniatures. Ed. by Roger C. Schank and Christopher K. Riesbeck. Hillsdale, NJ: Erlbaum, 1981. Pp. xii, 386. $29.95. Schank & Riesbeck present an introduction to one aspect of Artificial Intelligence (AI), namely the 'Yale view' of natural-language processing . They sketch the Yale AI Project's work in Chaps. 1-3; in Chap. 4 they describe LISP, the language in which their programs are written ; then they give their readers the chance to work with simplified versions ('miniatures') of five programs, and to learn first-hand what the programs do and how they do it. The book is written for 'those who want to comprehend how a large natural language-understanding program works' (p. ix); S&R therefore address computer scientists as well as psychologists and linguists, and in general they do an admirable job of balancing the book's content . Readers with some knowledge ofcomputer programming, even if they do not know LISP, will be able to follow the programs because of the contents of Chap. 4. Readers without such knowledge can skip some technical passages, but still follow the text to learn in a general way how the programs work. Each of the five miniature programs is preceded by a chapter which describes the fulllength program on which the miniature is based. These chapters are written by authors of the full-length programs, or by people especially familiar with them. Each miniature is followed by a set of exercises which guide the reader in extending and elaborating upon the program. The most basic of the five programs is the English Language Interpreter (ELI), which converts English sentences into representations called Conceptual Dependencies (CD's). The CD's present the meaning of a sentence as a set of roles, such as actor, action, object, and direction, which comprise an 'event'. ELI is unlike traditional parsing programs because it relies primarily on word meanings and memory, rather than syntax, to ???ßfGß? sentences. ELI reads a word and looks up its meaning, which tells what kinds of words should be expected to follow or precede the word being read, and how all the words will fit together into CD's. In cases of possible ambiguity, ELI stores words, and checks the context to determine which meaning is appropriate. S&R consider syntax as a tool in the understanding process, and they use principles of word order in ???ßfGe?3???; but they ignore questions of syntactic constructions . Unfortunately for linguists, there is little discussion ofthe ramifications which this theory of syntax has for future refinements in the programs. Short stories serve as input to three other programs: SAM, PAM, and POLITICS. First, ELI analyses the stories and produces series of CD's, which these programs ???ßfGß? in the light of 'scripts', which are memorized patterns of events for certain situations, e.g. going to a...