What is learned about fragments in artificial grammar learning? A transitional probabilities approach

Area of Science:

• Cognitive Science
• Psychology
• Artificial Intelligence

Background:

• Traditional models of sequential learning emphasize frequency-based encoding of regularities.
• The role of transitional probabilities in learning complex sequential structures remains underexplored.

Purpose of the Study:

• To investigate whether transitional probabilities, rather than chunk frequencies, are the primary drivers of artificial grammar learning (AGL).
• To propose and evaluate a transitional probability model (TPM) for encoding local regularities in sequential data.

Main Methods:

• A modified artificial grammar learning (AGL) procedure was employed.
• Participants were tasked with estimating the frequencies of bigrams within previously encountered exemplars.
• Sensitivity to transitional probability information versus pattern frequencies was assessed.

Main Results:

• Participants demonstrated a greater sensitivity to local transitional probability information.
• Mere pattern frequencies were found to be a less influential factor in learning.
• The proposed transitional probability model (TPM) offers an adaptive and parsimonious explanation for encoding sequential structure.

Conclusions:

• Transitional probabilities between elements are likely more critical than chunk frequencies for learning local regularities in sequential materials.
• The findings support the transitional probability model (TPM) as a key mechanism in artificial grammar learning (AGL).
• This research reframes the understanding of how the brain processes and learns sequential structures.