CEPA eprint 1331 (EVG-041)

Review of Schank’s “Conceptual Information Processing”

Glasersfeld E. von & Shumaker N. W. (1976) Review of Schank’s “Conceptual Information Processing”. Computers and the Humanities 10(4): 236–237. Available at http://cepa.info/1331
Conceptual Information Processing, Roger C. Schank. Amsterdam and New York: North-Holland Elsevier, 1975. Pp. vi + 374. $27.50.
When, at some future date, a History of Computational Linguistics is written, it might well open with the words: “In the beginning there was syntax.” The belief that formalized systems of syntactic analysis would enable computers to handle natural language dominated the field during the 1960’s and its influence is still noticeable in the neighboring areas of psycholinguistics and cognitive psychology. Short-lived though it was, it turned out to be one of the more expensive illusions of the post-war period. All along there were, of course, people who did not believe in the division between syntax and semantics and who persistently advocated a conceptual approach to language, e.g., Silvio Ceccato in Italy, Alfred Hoppe in Germany, Margaret Mastermari in England, and Roger Schank in the United States. What characterized these dissenters was, among other things, the idea heresy from the point of view of traditional linguistics that language is an instrument for the expression and communication of conceptual structures and that it must be studied together with the situational and cognitive contexts in which it is used.
In Conceptual Information Processing, Roger Schank and three of his students (Neff M. Goldman, Charles J. Rieger III, and Christopher K. Riesbeck) present a comprehensive summary of the ’conceptual dependency’ theory of language processing and its application to sentence comprehension, sentence generation, and a model of memory with inferential capabilities. Clearly that is a lot of subject matter for one book; but since the three applications are presented as illustrations of the theory rather than as self-serving descriptions of operational systems, the patient reader will be rewarded by a remarkably complete view of Schank’s theoretical edifice.
The communication-theoretical aspect of Schank’s basic assumptions, although he does not stress it, helps our understanding of his work if we remember that, quite generally, what is communicated is not meaning itself, but instructions to select meaning from a stock we have already stored. A piece of language, be it word, phrase, sentence, or paragraph, thus instructs the receiver to build up a ’conceptualization’ out of items and relationships with which he is already largely familiar owing to his prior linguistic and non-linguistic experience. Computational linguists have known for a long time that ’knowledge of the world’ plays a very important role in human language processing, but they have consistently avoided putting any such knowledge into their theories or into a computer. A computer could never handle it all, they have argued, because there is far too much; and it is impossible to make a selection, because one could never foretell which parts of it might be needed.
It is in this regard that Schank’s work is quite revolutionary. He refused to be misled by the traditional linguistic classifications of words (noun, verb, adjective, etc.) and went straight to the substrata of conceptual structures. There, rather than thousands of verbs, for instance, he found that a mere dozen ’primitive actions’ combined with some 15 conceptual categories were sufficient to characterize and graphically represent the meaning of the samples of English which he and his collaborators have so far examined. Since he is reporting on work in progress, not on a finished product, he explicitly states that these figures may not be definitive; but it seems safe to say that the total number of acts and categories needed to map our conceptual world is within the range of a few dozen.
It is important to stress that, unlike the transformationalist’s ’deep structure,’ Schank’s diagrams of conceptual structures are ’language-free’ and as such do not contain any indications as to the words or sentence structures a particular natural language may use to express them. This focus on conceptualization has led to a number of insights that should be of particular interest to semanticists.” To give one example, conceptual analysis shows that there are many verbs which, though they look like activities, do not indicate any particular activity but merely the fact that some unspecified activity causes a change of state. If, for instance, we hear “John has hurt Mary,” we merely know that he did something which caused Mary pain, but what he did remains open. This feature of activity situations (that they contain ’slots’ which may or may not be filled) is one of the keys to the conceptual parser. It is the as yet empty slots within the conceptual networks called up by single words that allow the parser to have ’expectations’ and to make certain inferences. At the same time, because these cognitive structures are networks, they contain the language user’s accumulated information as to which items are likely to play certain roles in an activity situation, which properties are likely to be attributed to certain items, and so on. The specification of relationships, which traditional linguists see as the task of syntax, is thus achieved by matching the linguistic input to experientially built-up networks and, roughly speaking, the best fit will be considered the most likely interpretation.
There can be little doubt that this approach to language data processing brings us a good deal closer to how the human language user proceeds. Though the exposition in this book is extremely compressed and the reader is likely to be left with some unanswered questions, the many clear and original insights make it a valuable contribution. From the user’s point of view, it is regrettable that no glossary or index of terms was included in a book whose function is expository, especially as the price seems high enough to include a bit of luxury.
Found a mistake? Contact corrections/at/cepa.infoDownloaded from http://cepa.info/1331 on 2016-05-13 · Publication curated by Alexander Riegler