Abstract
This paper, a collaboration between AI pioneer Patrick J. Hayes and psychologist Michael Leyton, explores how the "kinks" and "corners" of a shape can be used to reverse-engineer the physical processes that created it. By extending Leyton’s "process-grammar" to include sharp discontinuities, the authors demonstrate that the same rules governing smooth, organic shapes can also describe violent physical processes like crushing, pinching, and puncturing.
Introduction
In computer vision and cognitive science, a long-standing challenge is moving from a static geometric description of an object to an understanding of its history. In 1988, Michael Leyton proposed that human perception naturally interprets shape as the result of "processes"—for example, a protrusion is perceived as something having been "pushed out." While Leyton developed a grammar for smooth shapes, real-world objects often feature sharp creases and jagged edges. This paper addresses this gap by formalizing how these "first-order discontinuities" fit into the existing grammar of shape.
Theoretical Framework: Symmetry-Curvature Duality
The authors build upon the Symmetry-Curvature Duality Theorem, which states that every curvature extremum (a peak or valley in the curve) has a corresponding symmetry axis that points in the direction of the process that created it. To handle discontinuities, they model a "kink" as a point where curvature becomes infinite—a "curvature spike." They introduce an augmented notation that tracks spike parity (the direction the sharp bend points) alongside standard curvature extrema symbols ($M^+$, $m^-$, $m^+$, $M^-$).
The Grammar of Shape
The core of the methodology is the integration of these discontinuities into Leyton's original grammar. The authors show that operations such as continuation and bifurcation can be applied to "kinky" strings, allowing for complex causal chains to be represented symbolically.
As illustrated in Figure 2, the grammar provides a formal language for physical actions. For instance, "squashing" can evolve into "indentation" through continuation rules, while "protrusion" can be "squashed" or "resisted" through bifurcation rules. This allows for the representation of complex causal histories as sequences of symbolic transformations.
Key Insights: Pinching and Unified Grammar
A significant insight is the formalization of a cusp. Rather than a simple point, a cusp is described as the result of two curvature minima sliding toward each other and being "absorbed" by a central spike, providing a formal language for the act of "pinching."
The most significant contribution is the Unified Grammar: the original rules for smooth blobs still work; the system simply tracks the "spike" as an extra piece of metadata. This establishes that "kink-formation" is a natural extension of standard protrusion and indentation processes.
Conclusion and Future Work
The paper successfully demonstrates that complex, non-smooth shapes can be represented as strings of symbols and transformed using a small set of rewrite rules. However, the current notation assumes kinks occur exactly at curvature extrema, which limits the handling of "skew corners" or shapes with perfectly flat segments. Future work will focus on asymmetric phenomena and the automated extraction of these process strings from digital sensor data.
Summary updated