Back to results list
Please use this identifier to cite or link to this item:
|Title:||A linguistic approach to emotion detection and classification||Authors:||Lee, Yat Mei Sophia||Keywords:||Emotions -- Classification.
Language and emotions.
Hong Kong Polytechnic University -- Dissertations
|Issue Date:||2010||Publisher:||The Hong Kong Polytechnic University||Abstract:||Various linguistic and psychological theories of emotions have previously been proposed. However, none offers a concrete and comprehensive definition of emotion or can be used as the representational foundation for emotion analysis and processing. While it is often concluded that emotion concepts cannot be defined at all, this thesis argues that cause events, being the most tangible component of emotion, provide a rich dimension of how emotions should be classified. The current thesis sees emotion as a type of event which is triggered by actual or perceived events. It specifically focuses on the interaction between five primary emotions (HAPPINESS, SADNESS, FEAR, ANGER, and SURPRISE) and cause events. Cause events are examined in terms of two dimensions: namely transitivity and episternicity. I first present a qualitative analysis of cause events by examining the degree of transitivity based on three emotion-related features: agentivity, kinesis, and event participation, I observe that each emotion class has distinct cause event features. The five primary emotions can be put in a transitivity continuum where HAPPINESS and SADNESS occupy the two ends and the other three fall in between. I also examine the syntactic representation of cause events by identifying five types of epistemic markers heading the cause events based on a quantitative analysis of empirical data. These epistemic markers, SEEING, HEARING, KNOWING, DISCOVERY, and EXISTENCE, create a transparent environment for emotion causal relations. These linguistic findings reveal that the higher the experiencer's motivation to assert the certainty of the emotion, the more explicit epistemic marking of cause event is.
Based on the Natural Semantic Metalanguage model (NSM, Wierzbicka 1992) together with some insights from the Generative Lexicon (GL, Pustejovsky 1995), an emotion representation model combining event representation and emotion classification is proposed. Such representation assumes that emotion is decoded as an event type and is triggered by an event (cause event) and elicits another event (elicited event), By incorporating the semantic and syntactic information of emotion cause events, the emotion representation not only provides deep linguistic criteria of emotion cause events but also offers an event-based account of emotion classification. The linguistic account of emotion-cause interaction forms the basis of an automatic system to detect and classify emotion in text, First of all, an emotion cause corpus is created, which provides a valuable resource for both linguistic analysis as well as natural language processing of emotion and causes. Second, a text-driven, rule-based system to emotion cause detection is developed to attest to the validity of the linguistic model proposed in this thesis for emotion detection and classification. As a first step towards fully automatic inference of emotion-cause correlation, the rule-based cause detection system shows promising results, This thesis has some implications not only for the linguistic theory of emotions, but also for the linguistic account of events as well as the automatic detection and classification of emotion in language technology.
|Description:||xvi, 261 p. : ill. ; 30 cm.
PolyU Library Call No.: [THS] LG51 .H577P CBS 2010 Lee
|URI:||http://hdl.handle.net/10397/6369||Rights:||All rights reserved.|
|Appears in Collections:||Thesis|
Show full item record
Files in This Item:
|b26526542_link.htm||For PolyU Users||203 B||HTML||View/Open|
|b26526542_ir.pdf||For All Users (Non-printable)||9.47 MB||Adobe PDF||View/Open|
Citations as of Oct 22, 2018
Citations as of Oct 22, 2018
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.