Graph-based lemmatization of Turkish words by using morphological similarity

Lemmatization of the words is an important preprocess for Natural Language Processing (NLP) studies. Especially in language applications (such as part of speech tagging, spell-checking, and document clustering), selection of the right lemma with morphological features can provide better results. In this study, we present a new hybrid approach for Turkish inflected words by using morphological similarity based graph models which is recently getting popular in lemmatization. For this aim, a novel similarity function for Turkish is developed to connect the similar word forms. The proposed model is trained and tested by a double-checked Turkish lemmatization dataset. Then, empirical results are compared with ones of Zemberek which is the most used Turkish lemmatization tool.