Measuring Resilience using Language Modeling: A Computational Approach to Observing Resilience

Details

• Personal Author:
  Mahwish, Syeda ; Boyd, Ryan L. ; Varadarajan, Vasudha ; Kotov, Roman ; Luft, Benjamin J. ; Schwartz, H. Andrew ; Clouston, Sean A. P.
• Description:
  We developed resilience using language modeling (ReLM) to measure resilience in language through a novel natural language processing approach called archetype analysis. Our model conceptualizes resilience as a process of maintaining healthy functioning after an adverse event. ReLM is theoretically synthesized through nine facets of resilience reviewed from various sources as reflected in language that captures its dynamic capacity: optimism, sense of social support, emotional maturity, uncertainty tolerance, flexible mindset, coping toolkit, cognitive reappraisal, belief in a higher power, and continued activities of daily living. ReLM uses a language model to embed language in a semantic space, with cosine similarity to each facet's prototype statements calculated to quantify a theoretically derived facet score. We applied ReLM to 1,859 voicemails collected from 211 responders to the September 11, 2001, World Trade Center terrorist attacks. Principal component analysis on training and test sets identified a single latent factor from the facet scores, λ = 5.02 (56% variance explained), and measurement invariance testing confirmed scalar invariance across training and test subsets, deltax2(8) = 8.89, p = .352, indicating ReLM scores reflected the same underlying construct in both sets. A one-way analysis of variance showed significant differences in posttraumatic stress disorder (PTSD) symptom trajectories across resilience quartiles, F(3, 169) = 5.18, p = .002, with high resilience showing the largest improvements in PTSD after 4 years (M = -0.212). Using an archetype-based language model, ReLM offers a theoretically grounded approach to measuring resilience through natural language, capturing psychological processes in narratives, and enabling dynamic assessment.
• Subjects:
  Large Language Models Occupational Health Resilience, Psychological Safety
• Keywords:
  Post-traumatic stress disorder; PTSD; WTC; World Trade Center; Terrorism; Mental health; Health effects; Machine learning; Psychological processes
• ISSN:
  0894-9867
• Document Type:
  Text
• Funding:
  Cooperative Agreement
• Genre:
  Journal Article
• Place as Subject:
  New York ; Texas ; New York City Region [OSHA] ; Dallas Region [OSHA]
• CIO:
  National Institute for Occupational Safety and Health (NIOSH)
• Topic:
  Workplace Safety & Health
• Location:
  North America
• Pages in Document:
  14 pdf pages
• NIOSHTIC Number:
  nn:20071147
• Citation:
  J Trauma Stress 2026 Jan; :[Epub ahead of print]
• Email:
  Sean.Clouston@stonybrookmedicine.edu
• Federal Fiscal Year:
  2026
• Performing Organization:
  State University of New York - Stony Brook
• Peer Reviewed:
  True
• Start Date:
  20160901
• Source Full Name:
  Journal of Traumatic Stress
• End Date:
  20210831
• Collection(s):
  National Institute for Occupational Safety and Health
• Main Document Checksum:
  urn:sha-512:bb64fd879edb7705b901082e13c4811358b397d3c6ca78726b42591492b2cd8921e2e0a7f8e556e0de0db4b9fac6d65d50b15e4ae66cadaf91e0cf0fde339747
• Download URL:
  https://stacks.cdc.gov/view/cdc/257914/cdc_257914_DS1.pdf
• File Type:
  [PDF - 1.05 MB ]