Development of Morphological Analyzer for Sidaamu Afoo
No Thumbnail Available
Date
2020-06-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Sidaamu Afoo is one of the official working languages and mostly spoken in the Southern part of Ethiopia in Sidama National Regional State. As a working language, several documents are written by this language in this State. Yet, there is no any NLP applications that are tried before in this language. Analyzing these documents of this language solves several problems for the language. Morphological analyzer is a key task in natural language processing and concerned with how words are broken down into meaningful morphological units of language, and it serves as the base for other language. In our study supervised machine leaning by tuning memory-based learning approach is used. The system has two phases: training and analysis phases. In training phases, there is the process that languages morphophonemic process is analyzed, then morphologically annotated word is prepared. The features are automatically extracted by using windowing method from annotated words as a suitable to learn by learning model. While, analysis phase takes the surface word and analysis the morphophonemic process as a training phase, then features are extracted with fixed size windowing method. During extraction in analysis phase there is no need of specifying the class label as a training phase. Morpheme identification is processed after feature extraction and finally stems are extracted.
We have used Java to develop our system and TiMBL tool as learning model with IB1 and TRIBL algorithms for implementation. The experiments performance is evaluated by using 10-fold cross-validation and Leave-one-out techniques. The default and optimized parameter setting of algorithms is used to test the experiment. Finally, the system achieved good result of accuracy for both classifiers. The result of IB1 and TRIBL with default parameter setting is 93.65% and 96.03% respectively by using 10-fold cross-validation and by using leave-one-out IB1 achieved 93.4%. Also, with optimized parameter setting the IB1 and TRIBL achieved 96.83% and 97.03% respectively by using 10-fold cross-validation and by using leave-one-out technique IB1 achieved 96.6%. Thus, TRIBL algorithm achieved better performance than IB1 algorithm for Sidaamu Afoo morphology.
Description
Keywords
Sidaamu Afoo Morphology, Morphological Analyzer, Memory-Based Learning, Morphophonemic Process, Feature Extraction