Abstract:
Natural language is used to recode human knowledge. Data is stored in computers or on paper in
order to be processed and recorded for future use. Semantic role labeling (SRL) is one of the
essential problems in the field of natural language processing (NLP) and its task was to determine
the semantic roles (such as AGENT and PATIENT) of each argument that corresponds to each
predicate in a phrase automatically. SRL is useful shallow semantic representations, and it is a
very important intermediate step for several NLP applications, such as Information Extraction,
Question Answering and Machine Translation. Traditional methods for SRL are grounded on
parsing output, and require much feature engineering. So, the goal of this study was to develop a
semantic role labeller for Afaan Oromo text using a deep learning algorithm. To solve Afaan
Oromo SRL problems, we employ deep neural networks with long-short term memory (LSTM)
and bidirectional long-short term memory (BLSTM). For this study 1800, Afaan Oromo simple
sentences were used to train the system, which wassemantically annotated with semantic role label
using the BIO Tagging procedure and the PropBank annotation framework's principles. The
experiment conducted by using 90%, and 10% of the total dataset for training and testing
respectively. Experimental results show that the BiLSTM performed better and achieving the
better results in terms of accuracy (80%), precision (81%), recall (80%), and f-measure (80%) as
compared to LSTM in terms of accuracy (76%), precision (78%), recall (76%), and f-measure
(76%). Based on experimental analysis, concluding remarks and recommendations are forwarded
