Graphing Inline

Understanding Word-scale Graphics Use in Scientific Papers

Siyu Lu*Yanhan Liu*Shiyu XuRuishi ZouChen Ye

CHI EA 2026 · DOI 10.1145/3772363.3798356

* Equal contribution

PaperPreprintMaterial

Motivation

Example

At least 40 cities and counties are hiking their minimum wages: Flagstaff, Arizona, to $17.40; Mountain View, California, to $18.85; Denver, Colorado, to $18.29; and Portland, Maine, to $15.

Scientific figures typically sit apart from the text that cites them, so readers split attention between text and graphic — an extraneous load that can obscure the argument the figure is meant to support. Word-scale graphics offer a different path: compact visuals rendered at typographic size that fold the evidence back into the sentence itself. Prior work has mapped their design space, reading behaviors, and novel applications, yet how researchers actually put them to use in published papers remains uncharted. We therefore ask: How do researchers apply word-scale graphics in scientific papers?

Results

Our where–why–how framework and the corpus distribution across each dimension of word-scale graphics.

DimensionCategoryShareProportion
Where
Text+

Graphics that appear embedded in text descriptions of scientific papers.

65%
Table+

Graphics that appear embedded in tables of scientific papers.

35%
Why
Visual Indexing+

Graphics that serve the purpose of establishing a visual correspondence with a specific visual or textual entity mentioned in the text descriptions.

45.1%
Semantic Symbolizing+

Graphics that serve as visual aids to symbolize specific concepts, thereby reinforcing the text description to enhance reader comprehension.

38.1%
Data Annotation+

Graphics designed to allow direct extraction of data that directly encode quantitative data associated with texts.

16.8%
How
Icons+

Graphics that typically adopt stylized visual forms of objects designed to evoke semantic associations.

79.5%
Quantitative Graphs+

Graphics in which numeric data are encoded through certain visual channels (e.g., length, angle), usually containing an implicit coordinate system.

15.8%
Network Graphs+

Graphics that represent relational structures such as node-link diagrams to enhance the perception of relationships.

1.4%
Typography+

Graphics that encode texts with certain visual encodings (e.g., color) to create consistent visual linkages within the document scope.

3.2%

N = 909 word-scale graphics · Single-label coding per axis — percentages within each dimension sum to 100 · Corpus: 126,797 papers screened → 718 with word-scale graphics

BibTeX

Download
@inproceedings{lu2026graphinginline,
  author = {Lu, Siyu and Liu, Yanhan and Xu, Shiyu and Zou, Ruishi and Ye, Chen},
  title = {Graphing Inline: Understanding Word-scale Graphics Use in Scientific Papers},
  year = {2026},
  booktitle = {Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems},
  series = {CHI EA '26'},
  location = {Barcelona, Spain},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  numpages = {6},
  doi = {10.1145/3772363.3798356},
  url = {https://doi.org/10.1145/3772363.3798356}
}