The emotionality of lexicon in Cook Islands English online news

Abstract

Language – a means of communication used by humans within a society – enables people to interact to one another, to communicate emotions, and to develop emotions into specific categories. The relationship between language and emotion has been explored in different research fields, such as psychology, anthropology, linguistics, neuroscience. However, little corpus-based sociolinguistic research on how speakers convey their emotions in formal contexts has been carried out (Schweinberger, 2019). At present, there is little scientific consensus on the definition of emotion. From a psychological perspective, it is often assumed that emotion categories hold a universal essence; similarly, research on the emotion of lexicon argues that language is a representation of emotion categories which already exists (Lindquist & Barrett, 2008b). Whereas, anthropological research has revealed that the degree of emotional expressivity, verbally and non-verbally varies from culture to culture (cf. Wilce, 2009). This paper examines emotions which are conveyed by lexicon. Typically, emotions conveyed by words can be either explicit – represented by their core meaning – or implicit – represented by connotations. We first define the psychological construct of ‘emotion’, and refer to the Conceptual Act Theory (cf. Barrett, 2017), according to which basic elements represent sensations from inside (affect) and outside the body (e.g. vision), categories experienced in a given culture, and executive attention (Lindquist & Barrett, 2012). We employ sentiment analysis to investigate the social stratification of emotions. The sentiment analysis we refer to is based on the Word-Emotion Association Lexicon source (Mohammad and Turney, 2013), which comprises 10,170 lexical items. These words were assigned scores on the bases of ratings gathered through the Amazon Mechanical Turk service (see Schnoebelen & Kuperman 2010; Yu & Lee 2014; D’Onofio & Eckert, 2020 for discussion of the validity of use of Mechanical Turk for linguistic research). Commonly, most of the carried-out sentiment analysis only distinguishes between positive vs. negative emotions. Here, however, we explore the eight specific core emotions (anger, anticipation, fear, disgust, joy, sadness, surprise, and trust) in more detail by including them in the statistical model, which is described below. We built a small-scale corpus from web-based newspapers which contain leading articles, editorials, and letters to the editors from The Cook Island News and The Cook Islands Herald. For the quantitative analysis, we use a mixed-effects logistic regression model with a binary dependent variable: emotional vs. non-emotional lexical items. The independent variables include (a) type of emotion (positive vs. negative vs. neutral), (b) the eight core emotions, (c) class word (e.g. nous, adjectives, verbs), (d) sex of the author, (e) author, and (f) word-frequency. The latter is one of variables involved in language processing and is commonly studied in psycholinguistics, sociolinguistics, corpus linguistics (Mickiewicz, 2019). To account for word-frequency we use the SUBTLEX-US corpus (Brysbaert & New, 2009). Both ‘author’ and ‘word’ will be included in the statistical model as random intercepts.