Bad news or good news when recognizing emotional valence using phonemic content

Abstract

This study uses a psycholinguistics approach to the task of recognizing the emotional tone of texts with the purpose of providing an additional source of information related to emotional valence (positive-negative emotion). A previous analysis found that the rates of vowel-consonant pairings (“biphones”) are strongly correlated with emotional valence. Here we apply the discovered dependencies on arbitrarily downloaded texts of Internet news items. The sub-lexical level of these texts, with a length from 0.5 to 2.5 pages, is analyzed in regard to emotional tone. The result confirms that the phonemic level composed of biphones is indicative for the emotional tone of real-life texts. The developed method relies solely on the information carried by the sounds in the language (as an independent channel of information) and can be applied in systems for classifying the emotional valence of real-world texts.