Automatic sense prediction of explicit discourse connectives in Turkish with the help of centering theory and morphosyntactic features

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Enformatik Enstitüsü, Bilişsel Bilimler Anabilim Dalı, Türkiye

Tezin Onay Tarihi: 2018

Öğrenci: SAVAŞ ÇETİN

Danışman: DENİZ ZEYREK BOZŞAHİN

Özet:

Discourse connectives (and, but, however) are one of many means of keeping the discourse coherent. Discourse connectives are classified into groups based on their senses (expansion, contingency, etc.). They describe the semantic relationship of two discourse units. This study aims to build a machine learning system to predict the sense of explicit discourse connectives on the Turkish Discourse Bank data, which is manually gold-annotated. To do so, this study examines the effect of several features: i.e. transitions of Centering Theory and morphosyntactic characteristics of main verbs of the arguments in a discourse relation. The results imply that Centering Theory, morphosyntactic features and their combinations affect each class of sense in a different way. When the base score is calculated with only the connective feature, the addition of Centering Theory features seems to have increased the predictions scores for Comparison and Expansion classes. Also, Tense, Aspect and Modality features are observed to slightly affect the Temporal class in a positive way.