Toward More Effective Human Evaluation for Machine Translation
This work addresses the challenge of expensive human evaluation in machine translation for researchers and practitioners, offering an incremental improvement in efficiency.
The paper tackles the problem of reducing the cost and time of human evaluation for machine translation by using a sampling approach that leverages document membership and automatic metrics to predict scores for a complete test set, achieving up to 20% reduction in average absolute error compared to random sampling.
Improvements in text generation technologies such as machine translation have necessitated more costly and time-consuming human evaluation procedures to ensure an accurate signal. We investigate a simple way to reduce cost by reducing the number of text segments that must be annotated in order to accurately predict a score for a complete test set. Using a sampling approach, we demonstrate that information from document membership and automatic metrics can help improve estimates compared to a pure random sampling baseline. We achieve gains of up to 20% in average absolute error by leveraging stratified sampling and control variates. Our techniques can improve estimates made from a fixed annotation budget, are easy to implement, and can be applied to any problem with structure similar to the one we study.