POS Tagging for Amharic: A Machine Learning Approach

• Autor(a) principal: Kefena, Sintayehu Hirpassa
• Data de Publicação: 2020
• Outros Autores: Lehal, Gurpreet Singh
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: INFOCOMP: Jornal de Ciência da Computação
• Texto Completo: https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/627

In this paper, our focus is the problem of automatic prediction of Parts of Speech of words in Amharic language sentence. We present an experiment that involves the study and implementation of POS tagging model. Four statistical taggers, i.e. Trigrams’n’Tags (TnT) Tagger, Conditional Random Field taggers (CRF), Naive Bays (NB) and Decision Tree (DT) classifier is applying for a morphologically rich language: Amharic. We compare the performances of all taggers with the same size of training and testing Dataset. Various types of language-dependent and independent feature set have formed, and for each algorithm, a combination of them is applied. Based on such inputs CRF based model has achieved outperformed accuracy. The best accuracy obtained from our experiment is 94.08%. Finally, our study shows that linguistic features play a decisive part in overcoming the limitations of the baseline statistical model for Amharic languages.