Türkçe için doğal dil anlamada anlatım bozukluklarının tespiti için yeni bir yaklaşım geliştirilmesi

Defective expression is a grammatical term that refers both semantic and morphologic ambiguities in Turkish sentences. They are generally caused by misusing of a suffix in addition to absence or unnecessary use of an element in a sentence such as object, subject and etc. Having analyzed several studies related to this issue, it is found out that they are mostly performed by linguists by means of student questionnaires, tests or manual analysis by researchers. The absence of Natural Language Processing (NLP) studies related to this issue directed us to deal with this subject using computer technologies. However, grammatically demanding languages such as Turkish generally require rule-based and language-specific solutions especially in semantic problems. Rule-based systems have some major obstacles such as efficiency in processing, time consumption while development and intolerance for alteration in language. Machine learning models have made great advances in recent years, which led to unprecedented boost in NLP applications in terms of performance. In this thesis, we propose deep learning models of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) in addition to machine learning classifiers of k-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Random Forest (RF) to detect defective expressions in Turkish sentences. Experimental trials show that deep neural approaches come into prominence for detection in comparison to traditional classifiers. The study also reflects that due to having learning capability of long term dependencies, LSTM architecture will provide more promising results when amount of dataset is increased and more optimized. By being an original study in this field, this study is considered to make a great contribution to Turkish NLP and provides an excellent source for other researchers studying this area.