A Survey on Evaluation Metrics for Machine Translation

The success of Transformer architecture has seen increased interest in machine translation (MT). quality neural network-based MT transcends that translations derived using statistical methods. This growth research entailed the development accurate automatic evaluation metrics allow us to track performance MT. However, automatically evaluating and comparing systems is a challenging task. Several studies have shown traditional (e.g., BLEU, TER) show poor capturing semantic similarity between outputs human reference translations. To date, improve performance, various been proposed architecture. systematic comprehensive literature review on these still missing. Therefore, it necessary survey existing enable both established new researchers quickly understand trend over past few years. In this survey, we present metrics. better developments field, provide taxonomy Then, explain key contributions shortcomings addition, select representative from taxonomy, conduct experiments analyze related problems. Finally, discuss limitation current metric through experimentation our suggestions for further