A novel hybrid language model for finite state-based morphology

In this dissertation, we explore the combined usage of a rule-based approach and artificial neural network-based approach in Turkish morphological analysis. We design a language-independent novel hybrid algorithm combining the rule-based X-arbitrary morphological analyzer (XMOR) and an arbitrary ANN-based approach. The usage of hybrid models including both techniques is evaluated for performance improvements. Because of the agglutinative nature of the Turkish language, the suffixation of words is essential. A good rule-based morphological analyzer covers almost all words defined in the lexicon, but big data retrieved from freely available resources, such as social media, cannot be used as lexicon entries before correcting the error. Such words are usually erroneous and should be corrected prior to morphological analysis. The proposed hybrid approach is built on the idea of the dynamic generation of an artificial neural network according to two-level phonological rules. A combination of linguistic parsing, a neural network-based error correction model, and statistical filtering is utilized to increase the coverage of pure morphological analysis. The current hybrid method combines rule-based and long short-term memory-based techniques to increase the morphological analysis performance up to 99.90 percentage for OCRd data and 99.82 percentage for social media data, which represents a new state-of-the-art morphological analysis for Turkish to the best of the author's knowledge.