Reading Exception Words and Pseudowords: A Connectionist Network Simulation

A simulation experiment by David C. Plaut of Carnegie Mellon University demonstrates a connectionist network's ability to read words aloud. The network was trained on a 3000-word corpus and achieved over 99% accuracy on the training set, with performance on pronounceable nonwords comparable to human subjects. This work addresses the debate between dual-route and single-system models of reading.

Use Cases

• Benchmarking computational models of reading based on the described 3000-word training corpus.
• Investigating single-system versus dual-route theories of reading based on the described simulation results.
• Training connectionist networks for grapheme-to-phoneme mapping based on the methodology described.

Strengths

• Training accuracy of over 99% on the 3000-word corpus is reported.
• Performance on pronounceable nonwords is described as closely comparable to human subjects.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Carnegie Mellon University

Collection Method

Simulation experiments using a connectionist network trained on a specific corpus.