Text speaks louder: Insights into personality from natural language processing

In recent years, advancements in natural language processing (NLP) have enabled new approaches to personality assessment. This article presents an interdisciplinary investigation that leverages explainable AI techniques, particularly Integrated Gradients, to scrutinize NLP models' decision-making processes in personality assessment and verify their alignment with established personality theories. We compare the effectiveness of typological (MBTI) and dimensional (Big Five) models, utilizing the Essays and MBTI datasets. Our methodology applies log-odds ratio with Informative Dirichlet Prior (IDP) and fine-tuned transformer-based models (BERT and RoBERTa) to classify personality traits from textual data. Our results demonstrate moderate to high accuracy in personality prediction, with NLP models effectively identifying personality signals in text in line with previous studies. Our findings reveal theory-coherent patterns in language use associated with different personality traits, while highlighting important biases in the MBTI dataset that yielded less robust results. The study underscores the potential of NLP in enhancing personality psychology and emphasizes the need for further interdisciplinary research to fully realize the capabilities of these transparent technologies.