Using Neural Networks for Learning Arabic Morphophonological Rules
Abstract
Abstract
This study aims to investigate the extent to which artificial neural network models, specifically sequence-to-sequence (Seq2seq) models, can recognize Arabic morphophonological changes that occur to the lemma. The study focuses on four phonetic changes governed by the rules of Arabic verbal phonology. The results of this study reveal the ability of these models to learn phonological rules and recognize the contexts that constrain them. The model achieves a high accuracy when we test it on new datasets not encountered during the training process. Such results demonstrate the ability of sequence-to-sequence (Seq2seq) models to generalize over the contexts and learn phonological rules, despite the variation and the complexity in Arabic verbal system. This confirms what has been pointed out by previous studies conducted on morphological and phonological phenomena from other languages.
Keywords: phonetic changes, phonological rules, Arabic verb, neural networks, sequence models.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.