Idealised Cognitive Model (ICM) Analysis

Description

Explain why some cases are poor examples of a concept by stating the idealised model it presupposes, checking its fit, and splitting the concept into converging submodels and metonymic stand-ins.

Lakoff argues that words are defined against idealised cognitive models whose background assumptions do not fit the world perfectly. 'Bachelor' presupposes a society where men marry at a typical age, so the pope is a poor bachelor because the model does not apply, not because he partly belongs. 'Mother' is a cluster of birth, genetic, nurturance, marital and genealogical models; 'real mother' sentences, compounds such as stepmother and extended senses each pick out submodels. Metonymic models (stereotypes, typical examples, ideals, paragons, salient examples) let a subcategory stand for the whole and create further prototype effects, detectable with 'X, but Y' frames and inference asymmetries. Adapted for UX, the analysis explains why users mis-sort items, why a category label fails for edge cases, and which stereotype users reason from.

Attributes

Loss type
Latent
Effort
Days

Steps

  1. State the idealised model the term or category presupposes, including its background assumptions.
  2. Test candidate cases for how well that background fits, separating poor fit from partial membership.
  3. Where people's weightings conflict with their definitions, model the surrounding ICMs whose logic explains the weights.
  4. If no single model covers all uses, list the converging submodels.
  5. Probe submodels with 'real X' sentences, compound modifiers and extended senses in interviews or corpora.
  6. Identify metonymic stand-ins (stereotype, typical case, ideal, paragon, salient example) and test them with 'X, but Y' contrast frames and inference asymmetries.
  7. Summarise the composite prototype and the deviations it explains, and relate them to labels and categories in the product.

Sources

  • Women, Fire and Dangerous Things (Lakoff, 1987)

Also known as

  • ICM analysis
  • Background fit and cluster-model decomposition
  • Cluster model analysis
  • ICM fit analysis
  • Metonymic model analysis
  • Sources of prototype effects analysis

Categories

  • Content
  • Linguistics, Semiotics & Discourse