Sentence Level Amharic Word Sense Disambiguation

Abstract

Lexical ambiguity, phonological ambiguity, structural ambiguity, referential ambiguity, semantic ambiguity, and orthographic ambiguity were all types of Amharic ambiguity. The other ambiguities were out of this research because the study focuses on lexical-semantic, orthographic, and semantic ambiguities. Until now, some experts have been researching the Amharic word sense disambiguation system. Recent research, on the other hand, did not take into account antonym, troponymy, holonomy, and homonym relationships in the WordNet; this problem was overcome by this study. Using a Deep Learning method, we are developing an Amharic word sense disambiguation model. We use a design science research strategy to close the gap, starting with problem identification and concluding with final communication. 159 ambiguous words, 1214 synsets, and 2164 sentence datasets were used to create three distinct Deep Learning algorithms in three separate experiments. Using the given dataset, the overall performance of the model is measured using performance metrics in precision, F-measure, and confusion matrix. In this study, LSTM, CNN, and Bi-LSTM obtained 94 percent, 95 percent, and 96 percent accuracy respectively in the third experiment, based on performance measurement.