THE SYNTAX VERSUS SEMANTICS DEBATE REVISITED?

Glasgow’s thought-provoking article reminded me of the syntax versus semantics debates surrounding semantic nets and frames in the 1970s (Brachman and Levesque 1985). Her article is based on a fundamental distinction between descriptive and depictive representations, one which I do not believe can be drawn. Glasgow argues that representations such as arrays are depictive, as they are spatially analogical to the objects being represented, whereas representations such as propositional logic are descriptive, as they merely denote rather than analogically model the spatial entities of interest. This view, though, incorrectly attributes “depictiveness” to the syntax of a representation, whereas in fact it is a property of relationships defined on that representation. An array on its own is no more of an image analog than a set of propositions: “analog-ness” comes from the relations which we define between elements of the representation (e.g., “above”), and their correspondence to relations between objects in the real world. Arrays and propositions are thus no different; we can axiomatize spatial relations for both representations, and as a result they will both become more, but still equally, analogical to the original image. While there may be a difference in efficiency, Glasgow seems to be arguing more than this; that some representation languages are inherently “depictive,” whereas others are not. But “depictiveness” is conditional on the axioms or relations which the user chooses to define when using a language, be it propositional, array, or whatever. I emphasize this, as it is seductive to draw an array and decide it intuitively “looks” more like a spatial analog than a set of propositions. But this is a deception, created by the myriad of spatial relationships which the human mind overlays on the (much more impoverished) set of relations which a machine can compute, but which are subconsciously attributed to the machine nonetheless. In fact, all the machine “knows” are bytes (in physical locations bearing no correspondence to the locations of the objects they represent), and a small set of computations on these. Of course, there is still an efficiency issue: but this should not be cast as a debate between descriptive versus depictive representation languages but as between individual formalisms for spatial reasoning. Regarding the particular array formalism which Glasgow advocates, it is attractive in many respects and lends itself to parallel processing techniques. But at the same time treating space (an inherently continuous entity) as discrete, rectangular units may make reasoning about physical processes (also inherently continuous) problematic. While propositional logic may not be the answer, other formalisms for spatial reasoning have already been extensively studied in A1 (e.g., Forbus er al. 1991; Weld and deKleer 1990, Chapter 8), computer vision, computer-aided design, graphics, animation and simulation, which may offer more versatile and efficient solutions. A more important issue is whether spatial information should be preserved at all, and perhaps this is what is at the heart of the imagery debate. Psychological evidence indicates that mental imagery is fundamental to thought, and thus a contribution of the imagery debate to A1 is to reemphasize the importance of spatial reasoning. I hope, correspondingly, the computational perspective will highlight that depictiveness is not an intrinsic property of a representational structure, and thus that the question of descriptive versus depictive