The aim of this work is to define the categories GC, describe their categorical structure and show they are a model of Linear Logic. The second goal is to relate those categories to the Dialectica categories DC, cf.[DCJ, using different functors for the exponential “of course”. It is hoped that this categorical model of Linear Logic should help us to get a better understanding of the logic, which is, perhaps, the first non-intuitionistic constructive logic. This work is divided in two parts, each one with 3 sections. The first section shows that GC is a monoidal closed category and describes bifunctors for tensor “0”, internal horn “[—, —]“, par “u”, cartesian products “& “ and coproducts “s”. The second section defines linear negation as a contravariant functor obtained evaluating the internal horn bifunctor at a “dualising object”. The third section makes explicit the connections with Linear Logic, while the fourth introduces the comonads used to model the connective “of course”. Section 5 discusses some properties of these cornonads and finally section 6 makes the logical connections once more. This work grew out of suggestions of J.Y. Girard at the AMS-Conference on Categories, Logic and Computer Science in Boulder 1987, where I presented my earlier work on the Dialectica categories, hence the title. Still on the lines of given credit where it is due, I would like to say that Martin Hyland, under whose supervision this work was written, has been a continuous source of ideas and inspiration. Many heartfelt thanks to him. 1. The main definitions We start with a finitely complete category C. Then to describe GC say that its objects are relations on objects of C, that is monics A ~ U x X, which we usually write as (U ~ X). Given two such objects, (U ~- X) and (V L Y), which we call simply A and B, a morphism from A to B consists of a pair of maps in C, f: U — * V and F 4 Y —+ X, such that a pullback condition is satisfied, namely that where (~~)_1 represents puilbacks. (U x F) 1 (o~) ~ (f x Y) 1 (/3), (1) 342 Using diagrams, we say (f,F) is a morphism in GC if there is a (unique) map in ~, k: A ’ —~B ’ making the triangle commute: a~I Ia