Cross-Language Sentiment Analysis in Finance : Mapping English to French with Wordnet

This paper presents a novel technique to perform multilingual sentiment classification of financial documents drawing on methods from computational linguistics. The challenge is to efficiently analyze large volumes of text produced by corporations and companies in various languages to find features relevant for training machine learning algorythms. We use Open Multilingual WordNet, a large lexicon organizing words into semantic groups, called synsets, in more than 200 languages, as the basis for automatically translating sentiment words from English to other languages. We present experiments using a parallel English-French corpus of corporate annual reports. Results support the equivalence of measures across the two languages, showing consistent identification of positive and negative texts. We provide a detailed account of the method so that it can be extended to other languages and applied in machine learning