Deep lexical hypothesis: Identifying personality structure in natural language.

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Natural Language Processing

Background:

• Traditional psycholexical studies rely on survey-based ratings to understand adjective similarities.
• These methods are limited by sample size and the artificial nature of data collection.
• Recent advances in natural language processing (NLP) offer new possibilities for large-scale text analysis.

Purpose of the Study:

• To introduce and validate a method for extracting adjective similarities from language models using extensive, naturally occurring text.
• To compare the correlational structure derived from NLP methods with established psycholexical findings.
• To demonstrate the scalability and versatility of the NLP approach for semantic analysis.

Main Methods:

• Utilized general natural language processing (NLP) models to process millions of words from natural text.
• Extracted adjective similarities by analyzing co-occurrence and semantic relationships within the text data.
• Compared the resulting correlational structure with established psycholexical datasets, such as Saucier and Goldberg (1996a).

Main Results:

• The correlational structure derived from NLP closely matched traditional survey-based ratings.
• The first three unrotated factors from NLP analysis showed high congruence with established findings (coefficients of 0.89, 0.79, 0.79).
• The method's robustness was confirmed across different adjective sets, queries, and language models, although Neuroticism and Openness were less consistently recovered.

Conclusions:

• The NLP-based method provides a scalable and cost-effective alternative to traditional psycholexical studies.
• This approach offers a new source of semantic signal, aligning with the original lexical hypothesis.
• The method's applicability extends to multiple languages, historical texts, and very large datasets, facilitating broader research.