Unraveling the linguistic nature of specific autobiographical memories using a computerized classification algorithm

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psychopathology

Background:

• The Autobiographical Memory Test (AMT) is crucial for assessing memory dysfunctions, particularly in psychopathology.
• Distinguishing specific from nonspecific memories is vital for understanding memory recall.
• Linguistic analysis offers a potential avenue for objective memory classification.

Purpose of the Study:

• To develop and validate a computerized classifier for differentiating specific and nonspecific autobiographical memories.
• To identify universal linguistic features indicative of memory specificity.
• To assess the classifier's performance across different datasets and classification tasks.

Main Methods:

• Utilized a large Japanese corpus of cue-recalled memories (n=12,400) tagged for specificity.
• Extracted linguistic features such as tense, negation, and time/location adverbs.
• Trained a support vector machine (SVM) classifier and tested its robustness on novel memory data (n=8,478).
• Extended the classifier for a five-class categorization of memory specificity.

Main Results:

• The SVM achieved a high Area Under the Curve (AUC) of .92 in initial classification and .89 on novel memories.
• The classifier demonstrated robustness across different cue words and datasets.
• A five-class classification achieved 64%-65% accuracy, significantly above chance (20%).

Conclusions:

• Memory specificity can be reliably identified using a limited set of linguistic features.
• The developed classifier captures universal linguistic markers of memory specificity.
• This approach offers a computationally efficient method for memory assessment in research and clinical settings.