Representation Mapping: A Novel Approach to Generate High-Quality Multi-Lingual Emotion Lexicons

Representation Mapping: A Novel Approach to Generate High-Quality Multi-Lingual Emotion Lexicons Sven Buechel & Udo Hahn {sven.buechel|udo.hahn}@uni-jena.de Jena University Language and Information Engineering (JULIE) Lab F¨urstengraben 27, 07743 Jena, Germany http://www.julielab.de Abstract In the past years, sentiment analysis has increasingly shifted attention to representational frameworks more expressive than semantic polarity (being positive, negative or neutral). However, these richer formats (like Basic Emotions or Valence-Arousal-Dominance, and variants therefrom), rooted in psychological research, tend to proliferate the number of representation schemes for emotion encoding. Thus, a large amount of representationally incompatible emotion lexicons has been developed by various research groups adopting one or the other emotion representation format. As a consequence, the reusability of these resources decreases as does the comparability of systems using them. In this paper, we propose to solve this dilemma by methods and tools which map different representation formats onto each other for the sake of mutual compatibility and interoperability of language resources. We present the first large-scale investigation of such representation mappings for four typologically diverse languages and find evidence that our approach produces (near-)gold quality emotion lexicons, even in crosslingual settings. Finally, we use our models to create new lexicons for eight typologically diverse languages. Keywords: Automatic Construction of Emotion Lexicons, Representation Mapping, Models of Emotion 1. Introduction In the past two decades, the NLP-based analysis and pre- diction of affective states, as performed by sentiment anal- ysis systems, has received enormous interest (Liu, 2015). Starting with simple positive-negative polarity distinctions on the word or text level (Hatzivassiloglou and McKeown, 1997; Pang et al., 2002), research in sentiment analysis has since then shifted towards more nuanced and challeng- ing tasks, e.g., sentiment compositionality (Socher et al., 2013), aspect-level assessments (Schouten and Frasincar, 2016) or stance detection (Sobhani et al., 2016). In parallel, psychologically more advanced and more expressive repre- sentation formats for affective states have been proposed, like Basic Emotions (Ekman, 1992) or Valence-Arousal- Dominance (Bradley and Lang, 1994). However, there is currently no consensus in the literature what scheme should be used as a common ground. Rather, there are a multitude of competing formats often motivated by the needs of con- crete applications or the availability of user-labeled social media data (Desmet and Hoste, 2013; Li et al., 2016). While such decisions for a specific format may be perfectly reasonable in a specific research setting, on the flip-side, this proliferation of competing formats may seriously ham- per progress in sentiment analysis for two reasons, at least. First, language resources are less reusable (if at all) as gold standards and, second, with the growing number of repre- sentation formats meaningful comparisons between predic- tive systems become harder (if not impossible). One way to resolve this dilemma is to develop techniques to automatically translate between such formats. This task of emotion representation mapping (EMOMAP) was intro- duced only very recently to NLP by Buechel and Hahn (2017b). Their work came up with an emotion-labeled cor- pus which, in part, is annotated with two different emo- tion formats …

Content truncated to 500 words