Paper #11 - LADDER, A Sketching Language for User Interface Developers

Paper:
LADDER, A Sketching Language for User Interface Developers
(Tracy Hammond and Randall Davis)

Summary:
Shapes and groups of shapes, along with their actions of drawing, editing, and displaying, can be described with the sketch description language LADDER. This language allows interface designers a tool to handle these actions as well as make use of additional information such as stroke order or direction. The language incorporates predefined shapes, constraints, editing behaviors, display methods, and specification syntax for defining domains.

Describing shapes are limited to shapes with fixed graphical grammar, composed completely of LADDER primitive constraints, contain curves which are few in number or not highly relevant, and have high regularity. Shape definitions contain components, geometric constraints on those components, aliases for element simplification, editing behaviors, and display methods. Those definitions are simplified through a hierarchical structure. Shapes can be made abstract by requiring certain components to be defined. Shapes can also be made into a collection by having shapes in the same domain put into a group. The first advantage in grouping allowing the system top-down recognition, and the second is allowing shapes of the same group to move together. The language itself stems from various predefined shapes, constraints, editing tools, and display methods. In addition, components of variable numbers can be specified as a vector.

The LADDER recognition system is a multi-domain system which comprises of recognizing primitive shapes, domain shapes, and editing gestures. Low-level recognition is done on primitive shapes, and fact finding on a Jess rule-based system (Prolog-like syntax in Java) is performed on domain shapes. The system does checks on strokes to also determine if it is an editing gesture instead of a drawing gesture. Since shapes can be displayed with its original strokes, best-fit primitives, ideal strokes, or through Java Swing objects, the shapes’ components, properties, and constraints are converted into mathematical equations to find a global optima solution which satisfy all those equations. With new positional information obtained for description, the system creates and outputs a beautified shape.

Discussion:
Long discussed in one of his papers the lack of user interfaces employing sketch recognition due the difficulties interface designers face in employing them. Having a sketching language to alleviate these difficulties would facilitate the creation of more sketch-based interfaces, and the significance of LADDER lies in it being such a sketching language for interface designers. What makes LADDER even more significant is the fact that it is the very first sketch language. Shortcomings do exist, four of which are listed as limitations to shape design (those limitations are repeated in my summary above).

One aspect of LADDER that caught my attention was a particular facet of its multi-domain recognition system. Primitive shape recognition handles like it should, but the use of the Jess rule-based system in domain shape recognition was very intriguing. I heard claims that Jess was an expressive language, but I’ve only seen its applications in multi-agent systems. Recognition based on a rule-based system is a novel idea that I have not seen used in other system. I wonder if this has something to do with LADDER being a geometric recognizer as opposed to previous papers which focused on gesture recognition.