Connectionist Distributed Representation for Mental Model Theory

Description

Treats user mental models as graded similarity spaces and prototypes rather than explicit rule sets, eliciting them with similarity judgements and typicality ratings.

Classic mental-model research in HCI assumed users hold explicit, rule-like representations of how systems work. Connectionist theory, championed by Rumelhart, McClelland and the Churchlands, proposes that knowledge is distributed across weighted connections, so concepts behave like prototypes in a similarity space: membership is graded, typical examples are processed faster, and generalisation follows similarity rather than rules. Applied to UX, this suggests eliciting mental models through similarity judgements, typicality ratings and sorting tasks, then analysing them with multidimensional scaling or clustering, rather than asking users to state rules. Predictions follow: novel features are understood by their nearest prototype, and errors arise where a feature sits between prototypes. The approach faded as HCI moved away from cognitive theory.

Attributes

Loss type
Extinct
Effort
Days

Steps

  1. Instead of eliciting mental model rules ('users think the system works like X'), elicit mental model similarity spaces: present users with N interface elements, collect all pairwise similarity ratings, build a multidimensional scaling (MDS) solution.
  2. The MDS space IS the mental model — a geometric representation of how elements cluster and relate.
  3. Compare MDS spaces across user segments to reveal mental model differences.

Sources

  • Matter and Consciousness — Paul M. Churchland
  • Rumelhart & McClelland, Parallel Distributed Processing (1986)
  • Rosch, Principles of categorization (1978)
  • Cognitive Modeling: Nine Paradigms (slide deck)

Also known as

  • Similarity-space mental models
  • Prototype-based mental model analysis

Categories

  • User Research
  • Cognitive Psychology & Perception