Mining a crowdsourced dictionary to understand consistency and preference in word meanings

Abstract

Big data approaches to psychology have become increasing popular (Jones, 2017). Two of the main developments of this line of research is the advent of distributional models of semantics (e.g., Landauer and Dumais, 1997), which learn the meaning of words from large text corpora, and the collection of mega datasets of human behavior (e.g., The English lexicon project; Balota et al., 2007). The current article combines these two approaches, with the goal being to understand the consistency and preference that people have for word meanings. This was accomplished by mining a large amount of data from an online, crowdsourced dictionary and analyzing this data with a distributional model. Overall, it was found that even for words that are not an active part of the language environment, there is a large amount of consistency in the word meanings that different people have. Additionally, it was demonstrated that users of a language have strong preferences for word meanings, such that definitions to words that do not conform to people's conceptions are rejected by a community of language users. The results of this article provides insights into the cultural evolution of word meanings, and sheds light on alternative methodologies that can be used to understand lexical behavior.