Citation

Ahmad Azhar, Nur Syahmina and Mohd Zarifie Hashim, Nik and Iskandar Muhd Asramiza, Muhamad Amirul and Mat Ibrahim, Masrullizam and Abd Rahman, Noor Ziela and Dwi Sulistiyo, Mahmud (2025) AND Gate Perceptron for Word Prediction: A Comparative Study on Normal and Aphasic Communication. In: 2025 Multimedia University Engineering Conference, MECON 2025, 21 July 2025 - 23 July 2025, Cyberjaya, Malaysia.

Text IEEE Xplore Full-Text PDF_27.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Aphasia typically results from strokes and affects a person’s ability to communicate effectively. Although iconic and semantic gestures are capable of helping with speech problems, their computational alignment with lexical access is poorly understood. Traditional rehabilitation approaches are time-consuming and require in-person interactions with health professionals, which may limit accessibility. With the development of technology, deep learning offers an opportunity to build more effective assistive communication devices. This study presents the idea to develop a proposed model of an AND gate-conditioned perceptron-based model for better word prediction for normal and aphasic communication and, particularly, study its weight initialization strategies and their influence on word prediction accuracy. The proposed model achieved a higher degree of accuracy in word retrieval by focusing on these crucial deep learning components, such as activation functions, weight optimization, and loss function in the perceptron model, by combining lexical and gestural inputs through an AND gate mechanism. Experiments are done by initializing the fixed weights and also checking their stability and predicting the accuracy of normal and aphasic words, respectively. Findings indicate that the presented perceptron model can effectively assign stable weights following several training epochs when weight initialization is fixed and the weighted sum of all words exceeds the threshold value, resulting in the successful prediction of the word, which indicates its future use in deep learning rehabilitation systems. The fixed weight initialization improves lexical retrieval and helps in providing more stable and reliable word predictions to persons with aphasia. Combining this perceptron-based solution to the problem with the human-computer interaction concepts, the study suggests the possibility of more efficient and convenient aphasia rehabilitation systems.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Artificial Intelligence, Deep Learning
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	17 Mar 2026 07:03
Last Modified:	17 Mar 2026 07:29
URII:	http://shdl.mmu.edu.my/id/eprint/15517

Downloads

Edit (login required)