Cross-Linguistic Sentiment Analysis: From English to Spanish

International Conference RANLP 2009 - Borovets, Bulgaria, pages 50–54 Cross-Linguistic Sentiment Analysis: From English to Spanish Julian Brooke Department of Linguistics Simon Fraser University Burnaby, BC, Canada jab18@sfu.ca Milan Tofiloski School of Computing Science Simon Fraser University Burnaby, BC, Canada mta45@sfu.ca Maite Taboada Department of Linguistics Simon Fraser University Burnaby, BC, Canada mtaboada@sfu.ca Abstract We explore the adaptation of English resources and techniques for text sentiment analysis to a new language, Spanish. Our main focus is the modification of an existing English semantic orientation calculator and the building of dictionaries; however we also compare alternate approaches, including machine translation and Support Vector Machine classification. The results indicate that, although language- independent methods provide a decent baseline performance, there is also a significant cost to automation, and thus the best path to long-term improvement is through the inclusion of language-specific knowledge and resources. 1. Introduction Sentiment analysis refers to the automatic determination of subjectivity (whether a text is objective or subjective), polarity (positive or negative) and strength (strongly or weakly positive/negative). It is a growing field of research, especially given the gains to be obtained from mining opinions available online. Approaches to sentiment analysis have tackled the problem from two different angles: a word-based or semantic approach, or a machine learning (ML) approach. The word-based approach uses dictionaries of words tagged with their semantic orientation (SO), and calculates sentiment by aggregating the values of those present in a text or sentence [17]. The ML approach uses collections of texts that are known to express a favorable or unfavorable opinion as training data, and learns to recognize sentiment based on those examples [13]. Our approach is semantic, and makes use of a series of dictionaries, additionally taking into account the role of negation, intensification and irrealis expressions. We believe that a semantic approach offers the advantage of taking many different aspects of a text into account. One of the disadvantages of a semantic approach is that the resources necessary for a new domain or a new language need to be built from scratch, whereas a machine-learning approach only needs enough data to train. In this paper we show that porting to a new language, Spanish, requires only a small initial investment, while providing the opportunities for further improvement available only to semantic methods. For comparison, we have taken three approaches to performing sentiment analysis in a new language. Our main approach involves deploying Spanish-specific resources, which we build both manually and automatically. The second approach, used in Bautin et al. [4] and Wan [18], consists of translating the texts into English, and using an existing English calculator. Finally, the third approach builds unigram Support Vector Machine classifiers from our Spanish corpora. that, Our evaluation on multi-domain corpora indicates although translation and machine learning classification both perform reasonably well, there is a significant cost to automated translation. A language- specific SO Calculator with dictionaries built using words that actually appear in relevant texts gives the best performance, with significant potential for improvement. 2. …

Content truncated to 500 words