Detecting affective states from text based on a multi-component emotion model. (March 2016)

• Record Type: Journal Article
• Authors: Gao, Yingying; Zhu, Weibin
• Is Part Of: Computer speech & language. Volume 36(2016)
• Page Start: 42
• Page End: 57
• Publication Date: 2016-03
• Subjects: Multi-component emotion model -- Deep stacking network -- Visible intermediate layers -- Contextual impact -- Expressive speech synthesis; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2015.08.002
• Languages: English
• ISSNs: 0885-2308

Abstract

Highlights: A multi-component emotion describing model is proposed. A text-based emotion detecting method based on the describing model is proposed. Both the interactions among affective components and contextual effect are involved. Abstract: A multi-component emotion model is proposed to describe the affective states comprehensively and provide more details about emotion for the application of expressive speech synthesis. Four types of components from different perspectives – cognitive appraisal, psychological feeling, physical response and utterance manner are involved. And the interactions among them are also considered, by which the four components constitute a multi-layered structure. Based on the describing model, a detecting method is proposed to extract the affective states from text, as it is the requisite first step for an automatic generation of expressive synthetic speech. The deep stacking network is adopted and integrated with the hypothetic producing process of the four components, by which the intermediate layers of the network become visible and explicable. In addition, the affective states at document level and paragraph level are regarded as contextual features to extend available information for the emotion detection at sentence level. The effectiveness of the proposed method is validated through experiments. At sentence level, a 0.59 F -value of the predictions of utterance manner is achieved.