Do current connectionist learning models account for reading development in different languages?

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Linguistics

Background:

• Learning to read English, an irregular orthography, is demonstrably harder and takes longer than learning German, a regular orthography.
• Significant disparities in reading performance exist between children learning regular versus irregular orthographies, particularly in early stages of acquisition.

Purpose of the Study:

• To evaluate the capacity of contemporary connectionist learning models to simulate the observed cross-language differences in reading acquisition rates.
• To investigate whether computational models can replicate the slower learning trajectory for irregular orthographies like English compared to regular ones like German.

Main Methods:

• Implementation of German and English versions of two prominent connectionist reading models: Plaut et al.'s parallel distributed model and Zorzi et al.'s two-layer associative network.
• Assessment of model performance using nonword decoding accuracy as a measure of reading acquisition.

Main Results:

• Both implemented models predicted an overall advantage for the regular German orthography over the irregular English orthography.
• Neither model initially succeeded in simulating the pronounced early difference in learning rates between regular and irregular orthographies.

Conclusions:

• Current connectionist models require refinement to accurately simulate cross-language reading acquisition rates.
• Incorporating differences in pedagogical approaches, such as phonics versus whole-word methods, is crucial for models to capture real-world reading development patterns.