Browsing by Author "Abebe, Yobannes"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Development of Morphological Analyzer for Amharic Compound Words
    (Addis Ababa University, 2013-01) Abebe, Yobannes; Yimam, Prof. Baye
    The purpose of this study is to develop morphological analyzer for Amharic compound words. A number of researchers attempted to develop morphological analyzer for Amharic si nce 2000 (Abiyot, 2000; Tesfaye, 2002; Saba and Gibbon, 2005; Gasser, 20 II) and their analyzers provide a very good performance. However, as far as the researcher has noted, nothing is reported on their findings and results about analysis of compound words. For this reason, the researcher <. decided to develop morphological analyzer for Amharic compound words using rule-based approach on the basis of two-level morphology. A morphological analyzer is a computer program that takes a word or string of charcters as input and delivers an analysis as output. Amharic, in addition to simple words, uses compound words such as Me UO'}7 ... :f ayyar-mangadocc 'a irli nes', uo"h uo"l']9" malk-a-malkam ' beautiful', ,,)f7~~ laj -a-garad ' vi rgi n', etc. The developed anlyzer can recognize and deliver the given compound word with its word class, each constituents of the compound word with their POS and grammatical functions of the attached suffi xes. The study covers all compound categories in Amharic (i .e. compound nouns, adjectives, verbs, and adverbs) with their grammatical and syntacti cal information. The grammatical features included in this work are number, gender, person, case, and defi niteness. In identifying and analyzing compound nouns, adjectives, and adverbs, the system performs well and the sample used to the development and test set can be considered as representative of Amharic compounds. However regarding compound verbs, it covered only the main verbs, verb to 'say' and to 'do', and some of their variations, not all. In this study, algorithms that can identify and analyze Amharic compound words are developed from scratch. The performance of the system is evaluated using the training and test sets. The system accuracy on the test set is 98.67% and its precision and recall are 100% and 98.5%, respectively.