Enhancing Machine Translation of English Complex Sentences Using Refined Gradient CNN on Large-Scale Corpora

Abstract

Optimization of long and complicated sentences in English. Translating complex, lengthy statements from one language to another is the job of computer systems called machine translation algorithms (MTAs). A machine translation assistant (MTA) that trains on a big data corpus is one that makes use of a diverse and extensive collection of textual resources to improve translation quality. Translating complex and lengthy English sentences poses significant challenges for machine translation (MT) systems, especially when preserving semantic accuracy. It introduces the Refined Gradient-CNN (RG-CNN) model as a post-processing refinement mechanism to enhance phrase-level translation accuracy. The model is trained on a specially curated "Parallel Corpus" dataset comprising 1,563 English sentence pairs, including complex originals and their simplified counterparts. The RG-CNN employs gradient-enhanced convolution and bidirectional recurrent layers to capture and refine syntactic structures. The model is implemented using Python 3.11. Experimental results demonstrate the model’s superior performance. It achieved BLEU scores of 73.1% (corpus) and 70.1% (local), significantly outperforming. Likewise, RG-CNN reported a reduced WER of 0.3% (corpus) and 0.10% (local) compared to baseline models. Accuracy and recall were also improved to 97.51% and 98.43%, respectively, outperforming the baseline model. These results affirm RG-CNN's ability to optimize complex sentence translation, reduce ambiguities, and advance MT systems across diverse linguistic domains