SGL: A Structured Graphics Language

This paper introduces SGL, a graphics language that is aesthetically similar to SQL. As a graphical counterpart to SQL, SGL enables specification of statistical graphics within SQL query interfaces. SGL is based on a grammar of graphics that has been customized to support a SQL aesthetic. This paper presents the fundamental components of the SGL language alongside examples, and describes SGL’s underlying grammar of graphics via comparison to its closest predecessor, the layered grammar of graphics.

💡 Research Summary

The paper introduces SGL (Structured Graphics Language), a domain‑specific language designed to bring statistical graphics directly into SQL query interfaces. Built on Wilkinson’s Grammar of Graphics, SGL adopts a SQL‑like syntax—FROM, USING, VISUALIZE, GROUP BY, SCALE BY, LAYER, FACET BY, TITLE—so that users familiar with relational databases can specify visualizations without learning a new programming paradigm or JSON schema.

The authors first assume a data‑warehouse context with example tables (cars, trees). The FROM clause designates a single data source; multiple tables are disallowed, but a sub‑query can be supplied for preprocessing. USING selects geometric objects (geoms) such as points, line, bars, mirroring ggplot2’s terminology. VISUALIZE maps table columns to aesthetic attributes (x, y, color, size, etc.) and serves the role of SQL’s SELECT clause. Within VISUALIZE, column‑level transformations (e.g., bin, log) and aggregations (count, mean) are allowed, and a GROUP BY clause enforces the same rule as SQL: any non‑aggregated aesthetic must appear in GROUP BY. Importantly, scaling is performed before transformations, aggregations, and geom qualifiers, enabling visual operations (e.g., log‑scaled histograms) that are cumbersome in pure SQL.

The COLLECT BY clause extends GROUP BY to define how records are grouped into geometric objects. Individual geoms (points, lines) maintain a one‑to‑one mapping with post‑transformation records, whereas collective geoms (bars) map many records to a single object. This distinction underlies the “individual vs. collective” classification borrowed from ggplot2.

Geom qualifiers modify the positional behavior of geoms. Statistical qualifiers (e.g., regression) apply statistical transformations, while collision qualifiers (e.g., jittered) resolve over‑plotting. The LAYER operator allows multiple visual layers to be combined; all layers share a single scale per aesthetic and a single coordinate system, enforcing consistency across the graphic.

The SCALE BY clause explicitly sets scales (log, sqrt, etc.) for each aesthetic. Because scales are applied before any column‑level transformation, a regression line drawn after a log‑scale will be computed on the scaled values, preserving visual correctness. The coordinate system is inferred from the aesthetics used: x and y imply Cartesian, theta and r imply polar, enabling automatic generation of pie charts from stacked bar specifications.

FACET BY creates small multiples, partitioning the data by one or two expressions. An optional orientation keyword (vertical) controls panel layout. TITLE allows explicit axis labeling, overriding automatically derived titles.

Finally, the paper compares SGL’s grammar to the layered grammar of graphics, showing how defaults (dataset, mapping) are abstracted, how layers, facets, scales, and coordinate systems map between the two models, and how SGL simplifies specification by integrating these concepts into a single, SQL‑styled statement.

Overall, SGL offers a concise, expressive, and SQL‑consistent way to embed sophisticated statistical visualizations within relational query workflows, reducing the friction between data exploration and visual analysis and potentially streamlining the exploratory data analysis pipeline for database practitioners.