Semester of Graduation

Spring 2019


Master of Arts (MA)



Document Type



Schizophrenia is a clinical diagnosis assigned to individuals that experience positive (e.g., hallucinations and delusions), negative (e.g., blunted affect), and disorganized (e.g., incoherent speech) symptoms. One particularly disabling symptom is incoherence, which is defined as the meaning-based relationship between ideas. This symptom can drastically affect an individual’s quality of life by affecting areas such as social and occupational functioning. Currently, the mechanism behind this symptom is unknown and requires further study. One way to examine incoherence is to understand its level of expression in other clinical populations. With the advent of computationally-derived natural language processing (NLP), coherence can be quantified with more fine-grained detail at potentially lower levels of expression. Latent Semantic Analysis (LSA) is one promising methodology to examine coherence, but many unanswered technical questions about its application, specifically in clinical populations, still remain. Previous research has shown LSA can be used on speech from individuals with schizophrenia, who display the most extreme form of incoherence. To test LSA’s utility in other clinical populations and to specify parameters for its use, the current study used LSA on a “transdiagnostic” adult sample with varying forms of psychopathology. The current study aimed to extend previous findings in a different clinical sample and examine how coherence changes over time as a function of treatment. Results suggest that more traditional measures of coherence (i.e., clinician-ratings) were moderately correlated with LSA-measured coherence (r = 0.51). The optimal window size to differentiate high from low clinician-rated recalls was the entire recall, rather than eight words, as was previously found. Evidence for LSA-measured coherence’s dynamic nature was found as its reliability fell in the moderate range (a =0.72). This was close to clinician-rated coherence, with its reliability falling in the good range (a =0.79). Lastly, evidence supporting the incremental validity of LSA-measured coherence was not found as it was unable to provide unique variance in a model predicting clinical outcomes. Implications for these findings include additional evidence that newer computerized methodologies are related to traditional clinical measures and may provide insight into the dynamic nature of coherence.

Committee Chair

Cohen, Alex

Available for download on Monday, March 09, 2020