A Corpus-based Method for Extracting Paraphrases of Emotion Terms

Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 35–44, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics A Corpus-based Method for Extracting Paraphrases of Emotion Terms Fazel Keshtkar University of Ottawa Ottawa, ON, K1N 6N5, Canada akeshtka@site.uOttawa.ca Diana Inkpen University of Ottawa Ottawa, ON, K1N 6N5, Canada diana@site.uOttawa.ca Abstract Since paraphrasing is one of the crucial tasks in natural language understanding and gener- ation, this paper introduces a novel technique to extract paraphrases for emotion terms, from non-parallel corpora. We present a bootstrap- ping technique for identifying paraphrases, starting with a small number of seeds. Word- Net Affect emotion words are used as seeds. The bootstrapping approach learns extraction patterns for six classes of emotions. We use annotated blogs and other datasets as texts from which to extract paraphrases, based on the highest-scoring extraction patterns. The results include lexical and morpho-syntactic paraphrases, that we evaluate with human judges. 1 Introduction Paraphrases are different ways to express the same information. Algorithms to extract and automati- cally identify paraphrases are of interest from both linguistic and practical points of view. Many ma- jor challenges in Natural Language Processing ap- plications, for example multi-document summariza- tion, need to avoid repetitive information from the input documents. In Natural Language Genera- tion, paraphrasing is employed to create more var- ied and natural text. In our research, we ex- tract paraphrases for emotions, with the goal of us- ing them to automatically-generate emotional texts (such as friendly or hostile texts) for conversations between intelligent agents and characters in educa- tional games. Paraphrasing is applied to generate text with more variety. To our knowledge, most cur- rent applications manually collect paraphrases for specific applications, or they use lexical resources such as WordNet (Miller et al., 1993) to identify paraphrases. This paper introduces a novel method for ex- tracting paraphrases for emotions from texts. We focus on the six basic emotions proposed by Ek- man (1992): happiness, sadness, anger, disgust, surprise, and fear. We describe the construction of the paraphrases extractor. We also propose a k-window algorithm for selecting contexts that are used in the paraphrase extraction method. We automatically learn patterns that are able to extract the emotion paraphrases from corpora, starting with a set of seed words. We use data sets such as blogs and other annotated cor- pora, in which the emotions are marked. We use a large collection of non-parallel corpora which are described in Section 3. These corpora contain many instances of paraphrases different words to express the same emotion. An example of sentence fragments for one emotion class, happiness, is shown in Table 1. From them, the paraphrase pair that our method will extract is: "so happy to see" "very glad to visit". In the following sections, we give an overview of related work on paraphrasing in Section 2. In Sec- tion 3 we describe the datasets used in this work. We explain the details of our paraphrase extraction …

Content truncated to 500 words