Incremental Utility of Sentence Completion Test in Mental Health Assessment: Application of Natural Language Processing

The Sentence Completion Test (SCT) is a widely used tool in mental health practice to explore individuals’ psychological experiences and underlying vulnerabilities based on an open-ended response format. However, the attempts to objectively quantify and empirically validate them have been scarce. This study examined the utility of natural language processing (NLP) techniques in analyzing responses from the SCT for mental health assessment. Participants (n=535) completed SCT items and the Millon Clinical Multiaxial Inventory-IV (MCMI-IV). Transformer-based sentence embedding methods, including KoSBERT and KLUE-RoBERTa, were utilized. The NLP-based sentence embedding vectors derived from SCT responses effectively predicted MCMI-IV scores, indicating correlations above 0.30. Logistic regression analyses revealed that SCT-derived NLP scores significantly improved the prediction accuracy of identifying individuals currently undergoing mental health treatment, demonstrating incremental validity beyond traditional self-report measures. Findings suggest that brief SCT responses provide meaningful psychological information that complements self-report assessments.