Word Sense Disambiguation for Amharic Sentences using WordNet Hierarchy

Mulugeta, Mieraf

dc.contributor.author	Mulugeta, Mieraf
dc.date.accessioned	2020-03-16T09:13:25Z
dc.date.available	2020-03-16T09:13:25Z
dc.date.issued	2020-03-16
dc.identifier.uri	http://hdl.handle.net/123456789/10349
dc.description.abstract	Word Sense disambiguation (WSD) is an important application which can be integrated with different NLP applications for better performance. The presence of different types of ambiguities has been one of the main challenges for different researches and it is recommended to have the integration of WSD. Accordingly, though different attempts have been done to design Amharic WSD, there are problems on disambiguating all ambiguous words from an input sentence. The works done before can only disambiguate one target word at a time. Few studies also reported that WordNet is used as a knowledge base during the disambiguation process. However, the information contained in the WordNet and in the disambiguation is only definitions of the words, which is equivalent with dictionary based. On the other hand, when we see works which are corpus based, there is problem of knowledge acquisition and they are limited to only verb word class. Amharic WSD developed in this study is based on WordNet. Amharic ambiguous words used in the previous researchis used by adding relationships which are encoded in the WordNet and tested using augmented semantic space and context-to-gloss overlap implemented using python. Experiments are done to evaluate algorithmsimplemented in this study using Amharic sentences with ambiguous words. Word-level and sentence-level performance for one, two and three target words for different senses of ambiguous words are tested. Experimental results shows that, context-to-gloss followed by augmented semantic space has achieved the highest recall 87% and 79% for three target words at word and sentence level respectively. And the highest average accuracy 80% and 75% at word-level and sentencelevel is achieved by this approach. The major challenge in this study is getting data for both WordNet preparation and testing. The performance of the system can be increased if better stemmer or morphological analyzer is used, standard test sentences are used and fully constructed WordNet containing relationships for non-ambiguous words are used.	en_US
dc.language.iso	en	en_US
dc.subject	Computer Science	en_US
dc.title	Word Sense Disambiguation for Amharic Sentences using WordNet Hierarchy	en_US
dc.type	Thesis	en_US