Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Türkiye
Tezin Onay Tarihi: 2014
Tezin Dili: İngilizce
Öğrenci: Ahmet Faruk Acar
Danışman: DENİZ ZEYREK BOZŞAHİN
Özet:The subordinate verb forms that occur in non-finite adverbial clauses are called converbs (Göksel & Kerslake , 2005). In Turkish, converbs can be discourse connectives as well as acting as the complement of a factive verb or an adverbial. We morphologically analyzed 15 converbs in Turkish Discourse Bank to find out possible morpho-syntactic features in order to distinguish different roles of these converbs. The aim of the study is to find out all possible roles of the converbs and the source of ambiguities as well as to find out beneficial features that may promote automatic methods to disambiguate the discourse role of the converbs, namely Simplex subordinators. For this purpose, we created a converb-corpus out of Turkish Discourse Bank. We conducted an annotation experiment with two annotators and examined the results. Also we trained a decision tree algorithm to see whether the morphological features of the right and left material of the converbs are indicative for the disambiguation task. According to the annotation results, we observed three kinds of converbs: unambiguous converbs, which always create discourse relations; ambiguous converbs, which are ambiguous between a discourse connective and a non-discourse connective role; and hard cases, which are even more ambiguous, even for the human annotators. In addition to these, we saw that the syntactic features such as the syntactic class of the converb can be essential in automatic disambiguation studies. The distance between the converb and the matrix verb, and the morphological properties of the left and right edge of the converb seem to be good clues according to the machine learning experiment results.