Year: 2025
Type: article
Abstract: Machine translation is an important part of natural language processing, helping people communicate across languages, localise content, and search for information in different languages. In this article, we introduce a new framework using the Kolmogorov-Arnold Transformer (KAT) to improve translation quality. We test KAT on the Bilingual MELD dataset and compare it with both traditional statistical models and modern neural translation models. Our results show that KAT performs better, achieving a BLEU-4 score of 42.8, a METEOR score of 45.3, and a TER of 40.5, all of which are improvements over standard transformer models. We also find that using a larger vocabulary and adding the Kolmogorov-Arnold network helps improve translation accuracy. These results suggest that Kolmogorov-Arnold-based methods can be a valuable addition to machine translation systems.
Source: PeerJ Computer Science
Author: Yuzhe Nie
Institution: Shanghai University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,395
OpenAlex ID: W4414480730