Current methods of measuring and classifying depression, anxiety, and their shared cognitive process of perseverative thinking have significant limitations, being subject to demand and expectancy effects, cognitive biases, and reflexivity. Improved methods are needed that circumvent these limitations and reflect the transdiagnostic and dimensional nature of these constructs. The present research explored novel approaches for characterizing, measuring, and classifying these constructs. One method, natural language processing, holds potential for measuring internal states behaviorally. Chapter 1 carried out the first comprehensive examination of language features common and specific to depression and anxiety. Contrary to previous conclusions about the language of depression, many language features were shared between depression and anxiety. A subset of language features were relatively specific to depression (e.g., I-usage, sadness, and decreased positive emotion) or to anxiety (e.g., negative emotion, anxiety, and decreased negations), highlighting possible distinctions between these constructs. Chapter 2 used natural language processing to develop a behavioral measure of perseverative thinking. The covert process of perseverative thinking was associated with observable features of spoken language. A machine learning model based on these features accounted for meaningful variance in self-reported perseverative thinking; language-based perseverative thinking predicted clinically important outcomes including depression and anxiety, psychiatric comorbidity, and treatment seeking. Chapter 3 introduced a transdiagnostic, dimensional classification of anxiety as an alternative to the current classification system (DSM-5). The DSM anxiety classification is cumbersome and unaligned with evidence that a) anxiety problems cut across disorder categories and b) severity of anxiety matters, even at low levels. The novel transdiagnostic, dimensional classification offered meaningful gains in parsimony over the DSM anxiety disorders and showed noninferiority to DSM in predicting clinical outcomes. Collectively, these studies advance understanding of language features common and unique to depression and anxiety, demonstrate the feasibility of building language-based models of psychopathology constructs, and provide a methodological framework for the empirical evaluation of proposed transdiagnostic dimensional models of psychopathology. With further development, these novel methodological approaches could improve detection and facilitate treatment of depression, anxiety, and perseverative thinking.