ITU Validation Set for Metu-Sabancı Turkish Treebank

The Turkish Treebank (Oflazer et al., 2003; Atalay et al., 2003) created by the Middle East Technical University and Sabancı University is available to the researchers since 2003 and it is used by many researchers since then (Eryiğit and Oflazer, 2006; Eryiğit et al., 2006b; Eryiğit et al., 2006a; Nivre et al., 2007; Çakıcı and Baldridge, 2006; Buchholz and Marsi, 2006; Yüret, 2006; Wu et al., 2006; Dreyer et al., 2006; Shimizu, 2006; Schiehlen and Spranger, 2006; Riedel et al., 2006; Johansson and Nugues, 2006; McDonald et al., 2006; Liu et al., 2006; Chang et al., 2006; CorstonOliver and Aue, 2006; Cheng et al., 2006; Carreras et al., 2006; Canisius et al., 2006; Bick, 2006; Attardi, 2006; Eryiğit, 2006). Although it has some inconsistencies and still continues to be updated with newer versions, it served very much in the recent years for the development of the research on dependency parsing of Turkish. The Turkish treebank is composed of 5635 sentences and annotated with dependency structures. The modest data size of the treebank has been mentioned in many studies (Nivre et al., 2007; Buchholz and Marsi, 2006). There is no need to say that the size should be increased for better research on the field, but we should also state that the small size of the number of words (48K) of this treebank can be actually related to one of the features of the language itself. In the treebank, the average number of words in a sentence is 8.6 which is very lower when compared to other languages. This is since in Turkish, the words are sometimes equivalent to a whole sentence in another language which is a result of its agglutinative structure. This property of the language makes look the treebank smaller than it is when compared