MULTILINGUAL SENTIMENT NORMALIZATION FOR SCANDINAVIAN LANGUAGES

Abstract

In this paper, we address the challenge of multilingual sentiment analysis using a traditional lexicon and rule-based sentiment instrument that is tailored to capture sentiment patterns in a particular language. Focusing on a case study of three closely related Scandinavian languages (Danish, Norwegian, and Swedish) and using three tailored versions of VADER, we measure the relative degree of variation in valence using the OPUS corpus. We found that scores for Swedish are systematically skewed lower than Danish for translational pairs, and that scores for Norwegian are skewed higher for both other languages. We use a neural network to optimize the fit between Norwegian and Swedish respectively and Danish as the reference (target) language.