Briefly Noted

In a career spanning five decades, Patrick Suppes has pursued a unique approach to the problems of language and learning. This book, with Colleen Crangle, is a reprise of much of the work on the philosophy of language. As the title indicates, the underlying model is of a robot that is to receive instructions (and later also instruction) in ordinary language. The philosophizing is made concrete through software simulations and some simple experiments with a physical robot. Although much of the material has appeared before, it is reworked Into a cohesive whole. Any attempt to deal with such deep issues will involve simplifications, and it is instructive to follow the unfolding of complications. The starting point is quite conventional; attribute grammar with a semantics based on formal logic. But the demands of the robot domain quickly require the introduction of functional semantics for perception and action, which are not analyzed further, and for control primitives. Task requirements also push the authors to abandon the autonomy of syntax and to give a major role to intentionality, contextual readings, and the shared perceptions of speaker and hearer. Thus they are led to many of the core concerns of cognitive semantics and this becomes explicit in a chapter on geometry and spatial prepositions. Traditional issues concerning negation get extended in intriguing ways in the chapter that considers what it means to ask a robot to stop. The work on learning is mostly recent, although the theme goes back to Suppes's original work in mathematical psychology. The model has the robot choose parameters of actions (distance, etc.) from probability distributions that get modified by verbal feedback. The final chapter describes on-going work on the learning of natural language from examples in the robot domain. By postulating pre-existing internal structures for each utterance, the problem is reduced to statistically associating surface forms with internal constructs. This problem is analyzed theoretically and on a small corpus. As in the rest of the book, the work is idiosyncratic. But anyone with a deep interest in the philosophy of language can benefit from following the progression of ideas on some of its most fundamental issues.--Jerome A. Feldman, International Computer Science Institute, Berkeley