Relational Reasoning Underlies Creativity: Evidence From a Text-Mining Investigation

• In Event: 43.012 - Pushing the Boundaries of Relational Reasoning in Research and Practice: Cross-National, Cross-Domain, and Cross-Age Explorations

Abstract

Objective

This presentation demonstrates how the relational reasoning framework informs alternative assessment of divergent thinking by generating a text-mining algorithm that quantifies relational semantic distance among ideas.

Theoretical Framework

Within the educational and psychological literatures, it has been demonstrated that students vary on their ability to engage in higher-order reasoning, and that those differences predict cognitive performance (e.g., Dumas, Alexander, & Grossnickle, 2013). Whether those relations are conveyed via visuo-spatial stimuli (Dumas & Alexander, 2016), text (Alexander, Singer, Jablansky, & Hattan, 2016), or written language (Jablansky et al., 2015), RR ability appears to underlie convergent problem solving (i.e., converging on an optimal solution). Recently, studies have also suggested that RR plays a role in divergent problem solving, which entails generating multiple solutions to a given problem—an antecedent of creativity (Dumas, Schmidt, & Alexander, 2016).

For example, the most commonly utilized measure of divergent thinking ability is the Alternate Uses Task (AUT; Plucker & Makel, 2010). In this task, respondents are asked to generate as many novel uses for everyday object (cardboard box) as possible within a certain amount of time. The novelty or originality of a generated idea is determined by how closely related to the prompt that idea is. Ideas more closely associated with the prompt are less original, while ideas that are more distal are more original. This theoretical conceptualization opens the door for a conceptualization of divergent thinking based on RR in which RR is intrinsically linked to the production of original ideas, such as those generated to the AUT task (Hass, 2017).

Method

To investigate this hypothesized link between RR and originality, a reliable and valid measurement system is needed. Fortunately, a RR perspective on divergent thinking offers a solution: by quantifying the relational semantic distance among ideas, the originality of those ideas can be surmised. Here, we discuss the construction of a semantic algorithm trained on a massive corpus of text from the popular Internet site Reddit (over 1.6 billion separate Reddit comments are included in the corpus). Quantitatively, this algorithm formulates vectors within multivariate space that represent words or terms, and then measures the angle among those vectors to determine how closely associated the meanings of the words may be (Landauer, McNamara, Simon, & Kintsch, 2007). Using this semantic algorithm, more valid divergent thinking scores were achieved than were found for standard scoring methods (Dumas & Dunbar, 2014).

Significance

The reliability and validity of this text-mining system for the measurement of divergent thinking rests on the theoretical position that it is the relations among the ideas that define originality of ideas. In effect, the success of this text-mining approach to studying creativity was predicated on RR.

Authors

• Denis Dumas, University of Denver

• Peter Organisciak, University of Denver