Similarity and confidence in artificial grammar learning

Area of Science:

• Cognitive Psychology
• Artificial Grammar Learning
• Metacognition

Background:

• Artificial grammar learning (AGL) investigates how humans learn complex rule systems implicitly.
• Understanding the basis of confidence judgments in learning is crucial for cognitive models.
• Previous research linked objective similarity measures to categorization, but the role of subjective similarity in confidence was less clear.

Purpose of the Study:

• To investigate the relationship between similarity ratings and confidence ratings in artificial grammar learning.
• To determine if metacognitive confidence judgments are based on structural similarity knowledge.
• To explore how similarity influences confidence in grammatical decision-making.

Main Methods:

• Three experiments involving artificial grammar learning tasks.
• Participants provided similarity ratings of test items to study exemplars.
• Grammaticality decisions were made, followed by confidence or similarity ratings.
• Regression analyses and receiver operating characteristic (ROC) curve analyses were employed.

Main Results:

• Similarity ratings were related to objective similarity measures relevant to categorization.
• Grammaticality decisions were predicted by similarity ratings.
• Similarity and confidence ratings showed closely matched performance on ROC curves.
• Confidence ratings were influenced by the similarity of preceding items.

Conclusions:

• Metacognitive judgments of confidence in artificial grammar learning are predicated on structural knowledge of similarity.
• Subjective similarity is a key factor influencing confidence in grammatical decisions.
• Learners' confidence reflects their understanding of item similarity within the learned grammar.