dc.description.abstract |
Idiomatic expressions are a natural feature of all languages and are used frequently in our
daily conversations. Idioms are difficult to comprehend because their meaning cannot be
deduced immediately from the phrase from which they are derived. Idiomatic expressions
are phrases or clauses that have a common meaning that is unrelated to the individual
words' meanings. The existence of idiomatic expression in a text has an impact on
different natural language processing studies like Machine translation, sentiment analysis,
and semantics analysis are affected by idiom. Because idioms are one of the most
important aspects of a language, having an algorithm and mechanism for detecting them
is critical for NLP research. Several scholars conduct idiom recognition in different
languages however, there is not enough work on Amharic language idiomatic expression
identification. Models which are developed for other languages are not effective to
process Amharic documents due to morphological, syntactical, and semantic differences.
So, to address those gaps we developed a model for identifying the idiomatic expressions
from Amharic texts. This study used a deep learning approach to design an idiomatic
expression identification model and the corresponding idiomatic meaning for the
Amharic language using the Anaconda Jupiter notebook python framework. We have
experimented with CNN, LSTM, and BiLSTM deep learning algorithms using 4800
Amharic sentences for idiomatic expression identification. The accuracy of CNN, LSTM,
and BiLSTM was 94%, 95%, and 99% respectively. We compared the proposed model
with classical machine learning algorithms SVM and KNN. The experimental result
shows that the proposed BiLSTM model performs better than SVM, KNN, CNN, and
LSTM.
Keywords: Amharic Idiomatic Expression, Natural Language Processing, Deep Learning |
en_US |