This dissertation outlines a first-principles approach to automatically designing graphic presentations of information. The components of this approach include a conceptual framework for discussing how presentations encode information, algorithms for determining whether a method of presentation will be capable of presenting a given type of information, and design principles for ensuring the interpretability and perceptual effectiveness of a method of presentation.

Compared with previous approaches to automatically designing presentations, the approach outlined in this dissertation is more fine-grained and more general. It begins with an extremely general notion of how graphic presentations can encode information, then develops this into a useful framework by making a number of explicit assumptions about the types of presentations that people can use. This framework serves as a basis for analyzing the space of possible graphical languages -- i.e., the space of systematic methods of presenting data. The logical adequacy of different graphical languages for different types of information, and criteria and methods for composing graphical languages for different data are also explored.

In addition to this logical emphasis, this dissertation also emphasizes the influence of psychological issues on the design of presentations. It explores factors influencing the interpretability of presentations (i.e., how easily viewers will grasp how information is encoded) and outlines some general design principles for creating interpretable presentations. It also explores perceptual issues in presentation -- including perceptual organization, dimensional structure of visual stimuli, and the effectiveness of perceptual operations -- and outlines design principles for guaranteeing the perceptual effectiveness of presentations.

The last emphasis of this dissertation is on operationalizing the framework and principles -- i.e., on using them to create graphical languages in a relatively efficient manner. The implementation, AUTOGRAPH, demonstrates the flexibility and viability of a first-principles approach.