Validating Word Embedding as a Tool for the Psychological Sciences

Word Embedding (WE) represents a class of machine learning techniques used to model the relative meaning of words, based on the contextual interchangeability of those words within text corpora. Because the measurement of meaning is fundamental to the psychological sciences, psychologists have begun to adapt this tool from computational linguistics to study the human mind. While WE has been validated as a means to recover the meanings contained in text, this tool has not been successfully validated as a means to recover the meanings contained in our heads. Accordingly, in this dissertation I seek to determine the extent to which WE is applicable to psychological research. In Study 1, I investigate convergent validity between models of semantic memory based on word associations and models based on WE vectors, in three languages. Without parameterization, pre-trained WE vectors were found to recover more than 75% of the concepts (i.e., clusters) contained in word association networks. In Study 2, I explore the maximal degree of convergence between WE and word association models that can be achieved by applying different restrictions to WE models. At best, WE models were found to recover more than 95% of the concepts contained in word association networks. This level of accuracy was achieved when utilizing K-means clustering, focusing on specific parts-of-speech, and modelling concepts based on reliable word associations. In Study 3, I test whether psychological models can be recovered from WE vectors and probed. I found that the Big Five model of personality is largely recoverable from the personality descriptive space produced by WE, and that the Big Five trait factors function as attractors in this space. Implications for personality theory and research ensue. In the final chapter, I review the possible applications of WE modelling in the social and physical sciences by demonstrating how WE modelling can be used to answer broad questions regarding the content and nature of concepts. Taken together, my results justify WE as a means to probe the human mind, at a depth and scale that has been unattainable using self-report and observational methods.Ph.D