Türkçe geniş zaman ekinin öğrenimi: Çocuk edinimi ile morfolojik öğrenme modelinin karşılaştırılması

In this study, I apply a supervised morphological rule learner model, adapted from Albright and Hayes (2002), to test and improve the model's performance on the Turkish aorist by focusing on solving the problems posed by vowel harmony and irregular morphological patterns. The model generates rules based on word pairs differing in a single morphological feature. The research aims to evaluate the model's effectiveness by comparing its predictions with child language acquisition data from Nakipoğlu and Ketrez (2006). The study results show that the model captures the general patterns of vowel harmony and performs well on multisyllabic forms but struggles with monosyllabic verbs, especially those ending in sonorants. Additionally, I explore how syllable information affects the learnability of the aorist marker in Turkish and experiment with additional parameters. Overall, the study shows the limitations of rule-based approaches in modeling Turkish morphology and how syllable information helps in the learnability of the aorist marker. This study points out several avenues for future improvement, such as implementing corrective feedback, that could potentially change the results. The study hopefully has contributed to our understanding of the parallels between computational models and human language acquisition.