Sentiment translation for low resourcedlanguages: experiments on Irish general electionTweets

This paper presents two main methods of Sentiment Analysis (SA) of User-Generated Content for a low-resource language: Irish. The first method, automatic sentiment translation, applies existing English SA resources to both manually- and automatically-translated tweets. We obtained an accuracy of 70% using this approach. The second method involved the manual creation of an Irish-language sentiment lexicon: SentiFocloir. This lexicon was used to build the first Irish SA system, SentiFocalTweet, which produced superior results to the first method, with an accuracy of 76%. This demonstrates that translation from Irish to English has a minor effect on the preservation of sentiment; it is also shown that the SentiFocalTweet system is a successful baseline system for Irish sentiment analysis.