Dependency Based Amharic Grammar Checker
| dc.contributor.advisor | Assabie, Yaregal (PhD) | |
| dc.contributor.author | Gebreamlak, Abraham | |
| dc.date.accessioned | 2020-09-21T10:27:24Z | |
| dc.date.accessioned | 2023-11-29T04:05:55Z | |
| dc.date.available | 2020-09-21T10:27:24Z | |
| dc.date.available | 2023-11-29T04:05:55Z | |
| dc.date.issued | 2019-09-08 | |
| dc.description.abstract | Nowadays, advancement in computer and software technologies have reached the level of becoming basic necessity. The rise of computer-like gadgets such as smart phones, hand-held device, etc. is making the society paperless. As a result, we compose and exchange information on hourly basis. A grammar checker then come to play a role in identifying grammatical errors efficiently. An Amharic grammar checker was designed and developed through phrase structure grammar formalism which lacks to analyze nonlinear structure of a sentences. Another problems of phrase structure grammar is that as the complexity of phrases increases, it is difficult to treat directly through statistical model such as bigram and trigram models. These factors initiated us to design and develop a dependency based Amharic grammar checker. Accordingly, we propose an Amharic grammar checker integrated with dependency parsing system. The parser is based on dependency grammar formalism through which relationship of a word and its modifier is identified. The system is implemented through Python 3.7.2, UDpipe 1.0 to obtain and evaluate tokenizing and tagging models; MaltParser 1.9.2 to induce dependency parsing models and MaltEval 1.0 to evaluate the result of parsing model. The models were trained with a dependency treebank for Amharic. Lastly, we reported the performance of the induced models and grammar checker with randomly selected sentences from dependency treebank. The tokenizer and the tagger were also evaluated with raw texts collected from newspapers. We found out the tokenizer performs good with an accuracy of 100%. However the tagger’s performance was 43.11% on raw text corpus. The dependency parser were also evaluated during development to select best algorithm; which was Covington Non-projective algorithm. The agreement checker was evaluated for each agreement type. The results are 68.18%, 81.25% and 20% of subject-verb, object-verb, and adverb-verb agreements respectively. These results confirm us the feasibility of a dependency based grammar checker with an integrated dependency parser. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/22408 | |
| dc.language.iso | en | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Dependency and Phrase Structure Grammar | en_US |
| dc.subject | Dependency Parser | en_US |
| dc.title | Dependency Based Amharic Grammar Checker | en_US |
| dc.type | Thesis | en_US |