Türkçede ortaç yapılarına yönelik ekleme tercihlerinin göz izleme tekniğiyle incelenmesi

In this study, the processing of attachment preferences to relative clauses (RC) in Turkish was analyzed through an eye-tracking technique and comprehension questions presented following each experimental sentence. Within this general framework, the possible effects of the RC types (subject - object) on the processing of attachment types (low – high – high with ambiguity) and whether there was any RC asymmetry were examined. The data obtained from a total of sixty participants were analyzed in the study. Forty-two experimental sentences were developed based on two RC types and three attachment types as low, high and high with ambiguity. Therefore, in the study six conditions were tested along with forty-two filler sentences which were employed to distract the participants' attention away from the investigated structures. The Kolmogorow Smirnow test showed that the data exhibited a normal distribution. For two-way comparisons, an independent t-test was used and for three-way comparisons, the ANOVA (Analysis Of Variance) was employed. Certain differences between the two attachment types were observed. General direction of processing seemed to be that the High Attachment configuration caused slightly less cognitive load than the Low Attachment. However, the High Attachment sentences exhibited statistically significant longer reading durations on NP2 (the second noun phrase following the RC area). Therefore, it was assumed that the parser was sensitive to lexical/semantic properties of the incoming words of the given sentences during the initial processing. The processing of the main verb was another point of divergence. Significantly longer fixation durations on main verbs were observed in Low Attachment sentences, which also suggest that they included an implicit ambiguity. It was revealed that the main strategy of the parser to repair an ambiguity in High Attachment sentences was using the argument structure of the main verb. It acted as an error signal prompting the parser to adjust its syntactic preferences. It was found out that for all attachment types, on the RC Area of Interest (AoI) the object RCs are read with longer durations. However, statistically significant differences were not found. Considering the whole sentences, on the other hand, statistically significant results were found where participants spent longer durations for the processing of the object RCs. When analyzed alone, the subject RCs was comparatively easier for the parser to process than the object RCs. This is attributed to longer structural distance between the head noun and the extraction site besides linguistic-specific properties of Turkish in which the ORCs were inflected for person agreement, which caused extra processing load. Regarding the answers to the comprehension questions, the data complements the findings from online processing. The answers to the comprehension questions following High Attachment sentences had statistically the highest accuracy level. On the other hand, the comprehension questions following the High Attachment with Ambiguity sentences had statistically the lowest accuracy rate. Considering the answers to the comprehension questions concerning the subject and object RC sentences, it was observed that the comprehension questions concerning the subject RC sentences were statistically answered more successfully. It is suggested that For Turkish parser, the early processing is dominated by syntactic operations. However, it is overridden by lexical-semantic information of the main verb when it is led into a Garden-path situation. High Attachment sentences take the parser shorter to process compared to the Low Attachment sentences. Therefore, it is suggested that Turkish is a High Attachment language. Considering the reading times, Low Attachment sentences also include a local ambiguity as in High Attachment with Ambiguity. Finally, Cognitive load of ORCs is heavier than SRCs in all attachment types except for RC AoI. ORC asymmetry is also observed across all the experimental items regardless of attachment types, which is predicted by Structural Distance Hypothesis (SDH) (O'Grady, 2003).