A Classifier to Distinguish Between Cypriot Greek and Standard Modern Greek
Hanna Sababa, Athena Stassopoulou · 2018
The problem of discriminating between similar languages and dialects is one of the current challenges of natural language processing. In this paper, we describe the collection of a bidialectal corpus of Greek and the construction of a classifier to distinguish between Cypriot Greek (CG) and Standard Modern Greek (SMG). The corpus of CG and SMG was compiled from social media websites such as Facebook, Twitter and online forums. N-gram features were extracted and three classification algorithms were applied and tested on labeled sentences: multinomial naive Bayes (NB), linear support vector classifier (SVC) and logistic regression. All algorithms classified the test data with an accuracy of over 90%, with the multinomial NB classifier performing best, yielding a mean accuracy of 95%. This study adds to the existing body of work on the problem of discriminating between similar languages and is the first to examine CG and SMG. The results demonstrate the feasibility of an accurate Greek dialect classifier for academic or applied purposes.