For the purpose of hashtag recommendation, we designed and implemented a binary classifier, based on the Naive Bayes technique, which discriminates between English and non-English language tweets. Many approaches for language identification have already been proposed, but are often dependent on the type of content [6]. For each content type, manual labelling of a training set is needed when applying these supervised approaches. Therefore, we propose an approach which uses the Expectaton-Maximization (EM) algorithm to determine the parameters of Naive Bayes in an unsupervised manner. Given a dataset consisting of n observations x, let d be the number of features in an observation.