Refresh Step to Enhance Secure Translation Using Fully Homomorphic Encryption and Sequence-to-Sequence Neural Networks

Abstract

Secure translation has become crucial for governments and organizations to protect the exchange and translation of confidential documents and obtain accurate results in the target language. In a previous study, we discussed how combining fully homomorphic encryption and sequence-to-sequence neural networks could serve this purpose efficiently, which was demonstrated through its application and algorithm verification. The previous proposal presented limitations in terms of the amount of noise it produced, which affected the processing time, along with the limited database considered. Thus, a larger database in both Arabic and English needed to be considered while focusing on reducing time. Based on our previous work, this study found that adding a new “refresh” function to the steps decreases the noise from the multiple created layers; consequently, the computational time is reduced. Furthermore, the developed secure mode of the encryption process did not affect translation quality.