Labeling Emotion in Bengali Blog Corpus – A Fine Grained Tagging at Sentence Level
Dipankar Das, Sivaji Bandyopadhyay · 2010
Emotion, the private state of a human entity, is becoming an important topic in Natural Language Processing (NLP) with increasing use of search engines. The present task aims to manually annotate the sentences in a web based Bengali blog corpus with the emotional components such as emotional expression (word/phrase), intensity, associated holder and topic(s). Ekman’s six emotion classes (anger, disgust, fear, happy, sad and surprise) along with three types of intensities (high, general and low) are considered for the sentence level annotation. Presence of discourse markers, punctuation marks, negations, conjuncts, reduplication, rhetoric knowledge and especially emoticons play the contributory roles in the annotation process. Different types of fixed and relaxed strategies have been employed to measure the agreement of the sentential emotions, intensities, emotional holders and topics respectively. Experimental results for each emotion class at word level on a small set of the whole corpus have been found satisfactory. 1