Are baboons learning "orthographic" representations? Probably not.

Area of Science:

• Cognitive Science
• Neuroscience
• Computational Linguistics

Background:

• Baboons (Papio papio) and pigeons (Columba livia) can differentiate between English words and nonwords.
• Previous models focused on deep learning and ventral pathway simulations for baboon lexical decision-making.
• Species-specific differences in brain morphology suggest a need for a less specialized learning algorithm.

Purpose of the Study:

• To investigate if the Rescorla-Wagner learning model can explain word/nonword discrimination in baboons.
• To determine if a species-general learning algorithm underlies lexical decision tasks.
• To explore the role of low-level features versus explicit lexical representations in this behavior.

Main Methods:

• A discrimination learning network was employed, using gradient orientation features as input.
• The network's output units represented word and nonword categories.
• The model's predictions were compared against baboon lexical decision behavior data.

Main Results:

• The Rescorla-Wagner based network accurately predicted baboon lexical decision behavior, including similarity effects and learning curves.
• The model succeeded without explicit representation of words, relying on gradient orientation features.
• Performance demonstrated the model's ability to capture key aspects of the learning process.

Conclusions:

• The Rescorla-Wagner model provides a parsimonious explanation for word/nonword discrimination in baboons, suggesting a species-general learning mechanism.
• Baboons and pigeons appear to utilize low-level visual features and massively parallel processing, rather than explicit lexical hierarchies, for these tasks.
• Human reading may initially involve explicit letter learning but integrates with parallel, low-level feature processing for fluent reading, similar to non-human animal learning.