A minimal, runnable replication of the classic Interactive Activation (IA) model demonstration in McClelland & Rumelhart (1981), focusing on the “WORK” example.
- Simulator: Simbrain • Source: github.com/simbrain/simbrain
- Version used here: Simbrain 3.07, Mac (Apple Silicon)
- Workspace:
workspaces/IA(1981)_LetterPerception_WORK_demo.zip
Open this zip directly in Simbrain (no need to unzip).
- Install Simbrain
Download from the official site: https://simbrain.net
- Note: This demo was built and tested with Simbrain 3.07, Mac (Apple Silicon), but should also work with the corresponding Intel / Windows / Linux builds of 3.07.
- Open the workspace
- Start Simbrain
- Choose File → Open Workspace (.zip)
- Select: workspaces/IA(1981)_LetterPerception_WORK_demo.zip
- Run the simulation
- Press the Run / Play button to let the network iterate.
- Observe the dynamics
- At the word level, the node for WORK becomes strongly active, while competitors (WEAK, WEAR, GILL) remain suppressed.
- At the letter level, letters consistent with the word and position show higher activation, qualitatively similar to Fig. 6 in the paper.
All connection weights and parameters in this workspace strictly follow the values and descriptions given in McClelland & Rumelhart (1981).
-
All files in this repository (workspace, documentation, and images created for this demo) are released under the MIT License (see LICENSE).
-
Simbrain itself is distributed under its own license(s); please refer to:
-
Official website: https://simbrain.net
-
GitHub repository: https://github.com/simbrain/simbrain
-
If you use this demo in your teaching or research, please cite at least:
- McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88(5), 375–407.
and the Simbrain project:
- Simbrain project, https://simbrain.net