The article aims to discuss the problem of verbal features distinguishing emotionally neutral texts from emotional ones. Such features are necessary for elaborating a computer classifier designed to assign Internet texts in Russian to different emotional classes of texts including the neutral class. The complex analysis of the notion of neutrality in stylistics, axiology, emotiology and sentiment analysis allowed the author to formulate the preliminary hypothesis that emotionally neutral texts are marked by (1) the absence of obscene words and evaluative adjectives; (2) the non-expression of the modal frame of the utterance, which influences the functioning of xenomarkers (markers of narrative retelling), of deictic words and frequent predicates; (3) the lack of emotional vocabulary. The sources of the data for the study were three subcorpora: a subcorpus of texts from Wikipedia (136,184 tokens)—“neutral texts a priori”; a subcorpus of texts from the Russian social network VKontakte assessed by 2,000 informants as neutral ones—“neutral texts a posteriori” (14,000 tokens), and, finally, 15,000 emotionally loaded fragments of 60 to 80 words (1,092,327 tokens) from VKontakte assessed by the informants as texts manifesting different emotions: from anger to joy. The choice of the two neutral subcorpora was due to the author’s intention to examine the category of neutrality taking under consideration the genre of Internet texts. To learn the features of neutral Internet texts, the author used the methodology of corpus linguistics largely supported by a number of tools offered by the Sketch Engine corpus manager platform. The results partially confirmed the hypothesis and provided new details. Thus, the analysis revealed that the genre of the Internet text plays a crucial role in how the neutrality category manifests in it: for instance, Wikipedia texts have no obscene words, VKontakte texts assessed as neutral have a number of such lexemes. A similar discrepancy between neutral texts of different genres is observed on the grammatical level: for Vkontakte texts the collocation of the Russian verb byt’ [be] and the pronouns of the third person on, ona, oni [he, she, they] is statistically significant, for Wikipedia texts it is not. There are still some features proper to the two subcorpora considered as neutral: the absence of so-called “xenomarkers” and emotional vocabulary (words naming emotions, their manifestations, and typical situations in which people usually feel them); the low frequency, in comparison with emotional texts, of verbs and deictic words indicating time or place. The author has made a conclusion that, in the internet communicative space, the category of neutrality is construed differently than in others spheres of communication. Internet texts are generally so saturated with emotions that, even if in another context a text would be interpreted as an emotional one, in the social network users assess it as a fragment with no emotion. Another significant result is that the author has found a correlation between the type of narrator—narrator-experiencer, narratorobserver and narrator-non observer—and the type of text in terms of the category of neutrality. In further research, this will facilitate the predictability of text class—emotional or neutral.
Read full abstract