What if young people at risk for developing schizophrenia could be identified early, via a fast, automated, non-invasive test of language, which could be administered remotely? These youths could then receive intervention which might mitigate course and possibly prevent psychosis. Timed word fluency tests, in which individuals name words starting with a designated sound (typically F/A/S) or represent a given concept category (commonly animals/fruits/vegetables), have been used in the assessment of schizophrenia and its risk states, and in many other mental health conditions. Typically, psychologists manually record the number and size of valid phoneme clusters and switches observed in the phonemic tests and count the number of valid words belonging to a given category in the categorical tests. We present a new technique for automating the analysis of category fluency data and apply it to the problem of detecting youths at risk of developing schizophrenia, with best results over 85% accuracy when applying phonemic analysis to categorical data. The technique supports the separate quantification of structural and sequential phonemic similarity measures, supports an arbitrary range of pronunciations and dialects in the analysis, and may be extended to the assessment of other mental and physical health conditions, and their risk states.