Artificial Intelligence researchers and cognitive scientists commonly believe that thinking involves manipulating representations. Thinking involves search, inference, and making choices. This is how we model reasoning and what goes on in the brain is similar. Winograd and Flores present a radically different view, claiming that our knowledge is not represented in the brain at all, but rather consists of an unformalized shared background, from which we articulate representations in order to cope with new situations. In contrast, computer programs contain only pre-selected objects and properties, and there is no basis for moving beyond this initial formalization when breakdown occurs. Winograd and Flores provide convincing arguments with examples familiar to most artificial intelligence researchers. However, they significantly understate the role of representation in mediating intelligent behavior, specifically in the process of reflection, when representations are generated prior to physical action. Furthermore, they do not consider the practical benefits of expert systems and the extent of what can be accomplished. Nevertheless, the book is crisp and stimulating, and should make artificial intelligence researchers more cautious about what they are doing, more aware of the nature of formalization, and more open to alternative views.

A distinction is made between weak and strong research programs in cognitive science, the latter being characterized by an emphasis on the development of runnable computer programs. The paper focuses on the strong research program and initially considers situations in which it claims to have advanced our understanding of mathematical activity. It is concluded that the program’s characterization of students as environmentally driven systems leads to: (a) a treatment of mathematical activity in isolated, narrow, formal domains; (b) a failure to deal with relevance, common sense, and context, and (c) a separation of conceptual thought from sensory-motor action. Taken together, these conclusions imply a failure to deal adequately with the issue of mathematical meaning. In general, the program’s primary focus appears to be on programmable mechanisms rather than fundamental problems of mathematical cognition. The purview of the discussion is then widened to consider the strong program’s difficulties in dealing with social interaction, intellectual communities, and the hidden curriculum. It is noted that instructional implications derived from this program typically involve the organization of mathematical stimuli that make explicit or salient the relevant properties of a propositional mathematical environment. Finally, it is argued that some members of the strong program have recently acknowledged that it has limitations. The possibility of a rapprochement in which the strong program is supplanted by a form of social constructivism is discussed.

The second version of the Multistore Sentence Analysis System, implemented on an IBM 360/65, uses a correlational grammar to parse English sentences and displays the parsings as hierarchical syntactic structures comparable to tree diagrams. Since correlational syntax comprises much that is usually considered semantic information, the system demonstrates ways and means of resolving certain types of ambiguity that are frequent obstacles to univocal sentence analysis. Particular emphasis is given to the “significant address” method of programming which was developed to speed up the procedure (processing times, at present, are 0.5–1.5 seconds for sentences up to 16 words). By structuring an area of the central core in such a way that the individual location of bytes becomes significant, the shifting of information is avoided; the use of binary masks further simplifies the many operations of comparison required by the procedure. Samples of print-out illustrate some salient features of the system.

Key words: Computational Linguistics, Computer Programs, English, Form Classes (Languages), Kernel Sentences, Language Patterns, Machine Translation, Phrase Structure, Programing, Semantics, Sentence Structure, Structural Analysis, Structural Grammar, Syntax

The problem of representing information in automaton models of self-replication is considered. It is shown that, unlike in the natural reproduction process, in a computable model the reproduced entities do not contain all the information necessary for guiding the process. Current theoretical understanding of life and its replication, based on such models, is argued to be essentially inadequate. The solution to this problem is claimed to require recognition of the theoretical fact that information in living systems is different from that subsumed under the category of “knowledge”, which is representable as computer programs or triggers of state transitions. A discussion of fundamentals of a new theory of information and its relationship to replication models is given and a new direction of further developments of biological theories is envisioned.