Evaluation of a Non-Visual Molecule Browser
Andy Brown, Steve Pettifer, Robert Stevens · 2003 · Proceedings of the 6th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '04) · doi:10.1145/1028630.1028639
Summary
This paper evaluates Kekulé, software designed to allow visually impaired users to explore chemical molecular structures — diagrams that are typically presented as two-dimensional graph schematics showing atoms (nodes) connected by bonds (edges). Molecular diagrams are an important instance of the broader problem of making graphs accessible non-visually. Kekulé uses Chemical Markup Language (CML) as its data format and offers two complementary browsing modes: hierarchical browsing, which organises the molecule into a tree structure of functional groups (e.g., carboxylic acid, amine, phenyl) that users can zoom into and out of; and connection-based browsing, which allows navigation along bonds between individual atoms. A key innovation is that Kekulé automatically identifies and names functional groups and structural features (rings, chains) that are implicit in the raw graph structure but that sighted readers perceive at a glance through pattern recognition. The software uses speech output via the FreeTTS synthesiser, with keyboard commands for navigation, and provides orientation features including position reporting, hierarchy location, and component summaries. The authors draw on cognitive science research about why diagrams are useful — specifically Larkin and Simon's argument that diagrams group related information spatially, making search and inference easier — to argue that non-visual graph presentation must make these implicit spatial groupings explicit.
Key findings
Twelve sighted participants (justified on the grounds that sighted users without visual access face the same problems as naive blind users but without developed workaround strategies) compared the full version of Kekulé against a basic connection-only version across molecules of varying complexity. NASA-TLX scores showed the full version was significantly less demanding across mental demand (p=0.0016), frustration (p=0.0053), and effort (p=0.0107), with scores typically 2-3 points lower on a 20-point scale. Cooperative evaluation revealed that domain knowledge had a dramatic effect: a graduate biochemist answered one question in 20 seconds and 3 keystrokes using hierarchical browsing, while the same question took a typical participant approximately 90 keystrokes over 6 minutes in basic mode. Participants frequently lost orientation, particularly in cyclic structures (rings) where connection-based navigation could loop indefinitely. Several requested bookmarking or annotation features to mark visited locations. The naming of functional groups was found to have significant impact — it offloaded computational effort by providing meaningful labels that aligned with users' existing chemical knowledge. However, atom numbering based on CML file order rather than chemical convention caused confusion.
Relevance
This paper addresses the broader and persistent challenge of making diagrams and graph structures accessible to blind users — a problem that extends well beyond chemistry to UML diagrams, flowcharts, network maps, organisational charts, and any information presented as nodes and connections. The four principles extracted from this work remain highly relevant for accessible diagram design: (1) making implicit features explicit — automatically identifying and naming patterns that sighted users perceive through visual chunking; (2) enabling both hierarchical and connection-based browsing to support different exploration strategies; (3) allowing annotation and bookmarking to compensate for the lack of spatial memory that visual scanning provides; and (4) helping users maintain orientation within complex structures. The finding that domain knowledge dramatically accelerated exploration suggests that accessible STEM tools should leverage users' existing knowledge by presenting information using domain-standard terminology and organisation rather than raw structural data. For practitioners working on accessible data visualisations and diagrams, this work demonstrates that simply linearising visual content is insufficient — the implicit structure that makes diagrams useful must be computationally extracted and explicitly communicated.
Tags: STEM accessibility · non-visual interaction · blindness and low vision · diagram accessibility · synthesized speech · data visualization · graph accessibility · science education · accessible graphics