A Study on Evaluation Techniques for Machine Translation
Subhodeep Ghosh, Arin Ghose, Rishav Chattopadhya, Debranjan Sarkar · 2024
Effective evaluation of machine translation (MT) systems is critical for cross-lingual communication. Current methods like BLEU, METEOR, and BLEURT often fail to capture linguistic nuances. This paper examines the limitations of popular MT evaluation techniques and proposes two new methods to improve accuracy. We analyze current evaluation methods, identify their shortcomings, and develop RTN-Method and SAA-Method to address these issues. Our methods show significant improvements in capturing linguistic nuances and provide a more robust assessment of MT quality. The proposed methods offer a more reliable evaluation framework, enhancing the development of effective MT systems for better cross-lingual communication.