Differentiating dreaming and waking reports with automatic text analysis and Support Vector Machines.

Area of Science:

• Psychology
• Cognitive Science
• Computational Linguistics

Background:

• Understanding the linguistic differences between dream and waking states is crucial for cognitive and psychological research.
• Previous studies have explored dream content, but advanced computational methods offer new analytical possibilities.

Purpose of the Study:

• To compare word frequencies in dream and waking reports using Linguistic Inquiry and Word Count (LIWC) software.
• To investigate the applicability of machine learning for differentiating dream from waking reports based on linguistic features.

Main Methods:

• Analysis of word frequencies in dream and waking reports from Kahan and Sullivan (2012) using LIWC.
• Development and evaluation of a support vector machine (SVM) classifier trained on LIWC word counts for binary classification (dream vs. waking).

Main Results:

• Dream reports showed significantly higher frequencies of social words compared to waking reports.
• Positive emotion words were less frequent in dream reports than in waking reports.
• While overall cognition word counts were similar, specific cognitive process categories differed between states. The SVM achieved high performance metrics in distinguishing report types.

Conclusions:

• Findings support Social Simulation Theory (more social words in dreams) and Threat Simulation Theory (fewer positive emotions in dreams).
• Machine learning, particularly SVMs using LIWC data, is a viable and effective tool for automated dream and waking report classification.