Encoder: a connectionist model of how learning to visually encode fixated text images improves reading fluency.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Artificial Intelligence

Background:

• Reading fluency is crucial for comprehension.
• Current models of reading often simplify the visual encoding process.
• Understanding how the brain processes visual text information is key to improving reading.

Purpose of the Study:

• To investigate how visual encoding learning impacts reading fluency.
• To model the process of visual text recognition using a connectionist approach.
• To explore factors influencing the efficiency of visual encoding in reading.

Main Methods:

• Developed 'Encoder,' a connectionist model simulating visual text encoding.
• Trained Encoder on text images, varying image size and variability.
• Introduced regularities and biases in learning to mimic human reading patterns.

Main Results:

• Encoder's learning ability decreased with increasing image size.
• Reduced image variability and biased learning led to humanlike encoding accuracy.
• Trained Encoder demonstrated text familiarity effects similar to human readers.

Conclusions:

• Visual encoding learning can improve reading fluency by optimizing letter recognition.
• Model predictions align with computational learning theory regarding image size and variability.
• The study highlights the importance of sequential structure and learning biases in visual text processing.