Every tool encodes a theory of what visualization is.
The Grammar of Graphics, gave the visualisation field a shared vocabulary: data, aesthetics, marks, scales. That declarative grammar drove a generation of authoring tools: ggplot2, Tableau, and Vega-Lite, bringing rigour to chart-making and making it possible, for the first time, to talk about visualization in a common language. Arvind was part of building that future. Vega & Vega-Lite are his work.
That was a profound shift. And it came with a quiet cost.
The grammar that liberated visualization from fixed chart types also became its own kind of ceiling. The design space has surprising limits, inconsistencies, and cliffs. Seemingly simple charts like mosaics, waffles, and ribbons fall outside the scope of most implementations. The most expressive, unconventional visualization work often happens in spite of the grammar, not because of it.
Expanding what practitioners can make means expanding the grammar itself.
GoFish is Arvind's current answer to that problem: a declarative grammar built on Gestalt principles that enables recursive composition. Operators that can be nested and overlapped arbitrarily and opening a design space that the Grammar of Graphics could not reach. A step toward an updated theory of data visualization, one open to an infinite space of graphic representations rather than a finite set of design configurations.
The future is less settled. LLMs have entered the authoring loop. Generation is cheap. The boundary between the tool and the author is blurring in ways that could either democratize expressive visualization, or make us stop caring about expression altogether.
This keynote traces the full arc: the grammars that gave our field its foundations, the limits those grammars impose today, and the open question of what visualization authoring needs to become. What would authoring tools look like that keeps expanding rather than narrowing expressive range? And what do we lose if we let ease of generation replace intentionality of authoring?