Intelligent system for part-of-speech tagging using convolutional neural network on arabic language
Dhaya Eddine Messaoudi, Djamel Nessah, A. Siam · IET conference proceedings. · 2022
Part of speech tagging is the process of assigning the grammar and the morphology of words such as (noun, verb, adjective, article, and preposition... etc.). The principal challenges for this task are: first, the ambiguity (when a word can take several possible tags), and secondly the problem of rare words (in particular, words that did not appear in the training examples). In the past few years, the Neural Part of Speech Tagging has achieved competitive results with the use of deep learning, and this paper aims, thus, to provide an addition to the literature by the usage of Character-Level Convolution networks on Arabic language. We have experimentally implemented our model on the Boundary-Annotated Quran (BAQ) dataset, and we used the augmentation techniques to enhance the accuracy of the 1D convolutional neural network by adding handcrafted features. The results show 1D convolutional neural networks with a handcrafted feature outperformed, where we achieved a training accuracy of 97.04 % where the model without handcrafted features achieved 96.95 %.