|
Martin Johnson |
|
|
Associate Professor |
2222 Watkins Hall Tel 951.827.4612, Fax 951.827.3933 e-mail: martin.johnson@ucr.edu |
|
Computational Cognitive Politics
Meaning – how people understand symbolic representations of concepts – links the language, cognitive, and perceptual components of the language comprehension system. Curt Burgess (UC-Riverside Department of Psychology; Psycholinguistics and Computational Cognition Lab) has developed a tool for the study of the meanings that apply to words and concepts, called the Hyperspace Analog to Language memory model, or HAL. This computational approach analyzes language samples (in the form of text) and learns relationships among concepts much like a human language user. Given my own interests in opinion formation and the role of social context, Burgess and I are working on a variety of projects, mostly in collaboration with UC-Riverside Political Science graduate students Chad Murphy, Jackie Filla, and Tom Hayes, as well as Shaun Bowler. Our intent is to apply this tool to a variety of research questions, with the goals of assessing the generalizability of the HAL model to domains outside of Psychology and providing a new, rich measurement tool with accompanying high-dimensional language theory to the study of politics, rhetoric, and communication. Computational Context Analysis: A High-Dimensional Cognitive Modeling Complement to Content Analysis (with Chad Murphy and Curt Burgess) We present a high-dimensional theoretical model of cognitive representation, focused on concept acquisition. For students of large-corpus text analysis and text annotation, this model will be particularly interesting due to the computational instantiation it implies, which allows cognitive, political, and social researchers to systematically measure and study meaning in text. We develop replicable measures of cognitive belief structures, the semantic similarity of words, ambiguity, among other attributes of text-based expressions. The approach we describe complements the more traditional aims of content analysis, such as categorization. We discuss several applications of high-dimensional cognitive modeling to political science research questions, with a particular focus on priming racial stereotypes and the assessment of ideological positions in text. Strategies of Ambiguity: Modeling Rhetoric in Primary Election Campaigns (with Thomas Hayes, Chad Murphy, and Shaun Bowler) When politicians do not state clear policy positions, key steps in the process of democratic representation – accountability, responsibility and the idea of a mandate – are drawn into question. The strategic incentives for candidates to adopt ambiguous positions are, therefore, of wide normative importance. However, the limited empirical research in this area focuses on voter perceptions rather than the actual messages of the candidates. We test expectations that candidates’ semantic ambiguity will vary with their chance of winning the nomination, ideological extremity, and the campaign calendar. We test these hypotheses using a measure of ambiguity derived from a high-dimensional computational model of concept representation, the Hyperspace Analogue to Language (Lund & Burgess, 1996a). We collected textual data from presidential candidate websites in four waves between June 2007 and February 2008. We find little support for expectations informed by the current theoretical literature, but do find that Democratic presidential candidates discuss policy with more ambiguity than the Republican field. Investigating the roots of this difference, we find that Democratic voters are both more risk-acceptant and more ideologically diverse than Republican voters. Consequently, candidates in the major parties appear to be responding strategically to the characteristics of their voters during the primary season. |
Last updated April 10, 2008.