SemEval-2015 Task 9: CLIPEval Implicit Polarity of Events

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 443–450, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics SemEval-2015 Task 9: CLIPEval Implicit Polarity of Events Irene Russo ILC-CNR “A. Zampolli” Via G. Moruzzi, 1 56124 Pisa irene.russo@ilc.cnr.it Tommaso Caselli Vrije Universiteit Amsterdam De Boelelaan 1105 1081 HV Amsterdam t.caselli@gmail.com Carlo Strapparava Fondazione Bruno Kessler Via Sommarive, 18 38123 Povo (TN) strappa@fbk.eu Abstract Sentiment analysis tends to focus on the po- larity of words, combining their values to de- tect which portion of a text is opinionated. CLIPEval wants to promote a more holistic approach, looking at psychological researches that frame the connotations of words as the emotional values activated by them. The im- plicit polarity of events is just one aspect of connotative meaning and we address it with a task that is based on a dataset of sentences an- notated as instantiations of pleasant and un- pleasant events previously collected in psy- chological research as the ones on which hu- man judgments converge. 1 Introduction Current research in sentiment analysis (SA, hence- forth) is mostly focused on lexical resources that store polarity values. For bag-of-words ap- proaches the polarity of a text depends on the pres- ence/absence of a set of lexical items. This method- ology is successful to detect opinions about entities (such as reviews) but it shows mixed results when complex opinions about events - involving perspec- tives and points of view - are expressed. In terms of parts of speech involved, SA ap- proaches tend to focus on lexical items that explic- itly convey opinions - mainly adjectives, adverbs and several nouns - leaving verbs on the foreground. Improvements have been proposed by taking into ac- count syntax (Greene and Resnik 2009) and by in- vestigating the connotative polarity of words (Cam- bria et al., 2009; Akkaya et al., 2009, Balhaur et al., 2011; Russo et al. 2011; Cambria et al., 2012, Deng et al., 2013 among others). One of the key aspects of sentiment analysis, which has been only marginally tackled so far, is the identification of implicit polarity. By implicit polarity we refer to the recognition of subjective textual units where no polarity markers are present but still people are able to state whether the text portion under analy- sis expresses a positive or negative sentiment. Re- cently, methodologies trying to address this aspect have been developed, incorporating ideas from lin- guistic and psychological studies on the subjective aspects of linguistic expressions. Aiming at promoting a more holistic approach to sentiment analysis, combining the detection of implicit polarity with the expression of opinions on events, we propose CLIPEval, a task based on a dataset of events annotated as instantiations of pleasant and unpleasant events (PE/UPEs hence- forth) previously collected in psychological research as the ones that correlate with mood (both good and bad feelings) (Lewinsohn and Amenson, 1978; MacPhillamy and Lewinsohn, 1982). 2 Measuring Emotional Connotations: Psychological Studies For a long time research in psychology has been interested …

Content truncated to 500 words