Date of Award
Doctor of Philosophy (PhD)
Our understanding of Personality and its structure is rooted in linguistic studies operating under the assumptions made by the Lexical Hypothesis: personality characteristics that are important to a group of people will at some point be codified in their language, with the number of encoded representations of a personality characteristic indicating their importance. Qualitative and quantitative efforts in the dimension reduction of our lexicon throughout the mid-20th century have played a vital role in the field’s eventual arrival at the widely accepted Five Factor Model (FFM). However, there are a number of presently unresolved conflicts regarding the breadth and structure of this model (c.f., Hough, Oswald, & Ock, 2015). The present study sought to address such issues through previously unavailable language modeling techniques. The Distributional Semantic Hypothesis (DSH) argues that the meaning of words may be formed through some function of their co-occurrence with other words. There is evidence that DSH-based techniques are cognitively valid, serving as a proxy for learned associations between stimuli (Günther et al., 2019). Given that Personality is often measured through self-report surveys, the present study proposed that a Personality measure be created directly from this source data, using large pre-trained Transformers (a type of neural network that is adept at encoding and decoding semantic representations from natural language). An inventory was constructed, administered, and response data was analyzed using partial correlation networks. This exploratory study identifies differences in the internal structure of trait-domains, while simultaneously demonstrating a quantitative approach to item creation and survey development.
Moore, Alexander, "“I Can See the Forest for the Trees”: Examining Personality Traits with Trasformers" (2022). All Dissertations. 3055.