Can Computational Linguistics Help Explain Political Trust?
Eduardo Araral, National University of Singapore, Singapore
The theoretical literature on trust suggests that its variations can be explained by the structure of the game i.e. group size, history of cooperation, face to face communication, repeated interaction, norms, among others. The survey literature on the other hand ask respondents whether and to what degree they trust a politician or an institution. In both cases, little is understood how trust is defined by participants or respondents, in large part because of cost limitations. In this paper, we illustrate a cost effective and complementary method to study political trust using computational linguistics or semantics analysis - the study of the linguistic meanings of morphemes, words, heuristics, phrases, and sentences. Based on the folk wisdom that the meaning of a word can be understood from the company it keeps, semantics analysis is important because our reality and behavior is framed by words, heuristics and phrases. We used the Corpus of Historical American English (CoHA) database – comprising 400 million words - and divided it into decade-long segments, stretching from 1820 to 2020. For each segment, we used a Word2Vec model to identify the words with the closest vector distance to "trust" thus providing a map of how the concept has evolved over time. We then tested whether there is a correlation between vector proximity of election candidates’ names to trust-related words and their political success. We find support for our hypotheses. Finally, we reflect on the potentials and limitations of this method for political science research and suggest areas for future research.
Presentation Date/Time: Sunday, December 11, 2022 (11:20)
Session: Session 2
Room: Fai Kham Room