eGRUMET: Enhanced Gated Recurrent Unit Machine for English to Kannada lingual Translation
Sumukh R Kashi, H R Vineeth, H S Gururaja · 2023
A branch of natural language processing called neural machine translation (NMT) focuses on using artificial neural networks for translating textual content in one language to another, preserving the semantics of the source language. The ability to provide high-quality translations that are frequently comparable to those generated by human translators is a cogent need and therefore, NMT systems have gained popularity. This article proposes a novel gated recurrent unit (GRU) machine that is functionally enhanced and tailored for English to Kannada lingual translation. The input data is first pre-processed by removing stop words, followed by the process of stemming and lemmatization. The approach described in this paper compares a primitive GRU system with the proposed system that is structurally denser and efficient. The novelty of the translation system has been achieved by two newly introduced blocks in the attention layer of the GRU machine. The BiLingual Evaluation Understudy (BLEU) performance evaluation of the enhanced GRU machine is performed on a dataset composed of English text with corresponding semantically equivalent Kannada text in order to test the efficacy of the method. The performance metric of time taken for translation of text from English to Kannada has been computed for a basic GRU system and the proposed system on multiple sentence-pairs. Both performance measures indicate better values with the proposed method.