Effect Functors for Opinion Inference

Effect Functors for Opinion Inference Josef Ruppenhofer, Jasper Brandes Universit¨at Hildesheim Universit¨atsplatz 1, 31141 Hildesheim {ruppenho, brandesj}@uni-hildesheim.de Abstract Sentiment analysis has so far focused on the detection of explicit opinions. However, of late implicit opinions have received broader attention, the key idea being that the evaluation of an event type by a speaker depends on how the participants in the event are valued and how the event itself affects the participants. We present an annotation scheme for adding relevant information, couched in terms of so-called effect functors, to German lexical items. Our scheme synthesizes and extends previous proposals. We report on an inter-annotator agreement study. We also present results of a crowdsourcing experiment to test the utility of some known and some new functors for opinion inference where, unlike in previous work, subjects are asked to reason from event evaluation to participant evaluation. Keywords: Opinion Inference, Effects, Subjectivity 1. Introduction So far, sentiment analysis has mainly focused on the detec- tion of explicit opinions. However, recently the relevance of implicit opinions has received broader attention within the field. Here, the work of Klenner and colleagues on verb polarity frames (Klenner et al., 2014; Klenner, 2015), the functor-based framework of event evaluation by Anand and Reschke (Anand and Reschke, 2010; Reschke and Anand, 2011) and the work on opinion implicature by Wiebe and her colleagues (Deng et al., 2013; Deng and Wiebe, 2014; Wiebe and Deng, 2014), based on the effects that events have on object entities, have been the most important con- tributions to the theory of implicit opinions. The key observation of this line of work is that, while some predicates do not denote sentiment, in combination with expressions of explicit sentiment and knowledge about atti- tudes towards entities, one can infer implicit opinions being conveyed that hinge on these inference-relevant predicates. Consider (1). From the explicit positive sentiment (in bold- face) by Peter towards the fact that the Colts lost, an event which affects them negatively (in SMALL CAPS), one can infer that he has a negative sentiment towards the Colts. Furthermore, depending on the context, one may infer that the speaker/writer producing (1) may have a positive or negative attitude towards Peter, depending on whether the speaker disapproves or approves of the Colts. (1) Peter is happy because the Colts LOST. While we expect a sentiment lexicon to provide us with in- formation that e.g. happy expresses positive sentiment, the information that the effect-bearing word lose entails nega- tive affect on the subject can so far not be retrieved from any lexical-semantic resource. In this contribution, we push the effort begun by Ruppen- hofer and Brandes (2015) toward a synthesis between the effect-based approach and the functor approach further in several respects. Firstly, we identify and propose additional functors for verbs embedding states of possibility, predi- cates expressing location, and predicates expressing simi- larity. These functors allow us to cover predicates whose entailments have so far not been allowed for by Anand and Reschke’s functors (Anand and …

Content truncated to 500 words