The language of accurate recognition memory.

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Machine Learning

Background:

• Recognition memory research traditionally focuses on accuracy and response times.
• The rich linguistic data accompanying memory judgments remains underexplored.
• Understanding the basis of recognition decisions is crucial for cognitive science.

Purpose of the Study:

• To investigate if natural language used in memory justifications predicts recognition accuracy.
• To explore the potential of machine learning and receiver operating characteristics (ROCs) in analyzing memory justifications.
• To determine if linguistic cues reflect underlying cognitive processes in recognition memory.

Main Methods:

• Utilized machine learning classifiers trained on linguistic features of memory justifications.
• Employed receiver operating characteristics (ROCs) to analyze decision accuracy.
• Tested the generalizability of findings across different experimental procedures.

Main Results:

• Natural language justifications accurately predicted 'old' (hit vs. false alarm) decisions, but not 'new' decisions.
• The linguistic classifier outperformed decision speed in predicting accuracy.
• The classifier generalized to independent datasets and showed sensitivity to reported recollective experiences.

Conclusions:

• Memory justifications provide a valuable, untapped source of information about recognition memory processes.
• Linguistic analysis can reveal explicit bases for classification decisions in cognitive tasks.
• This approach has potential applications in basic cognitive science and applied fields.