Linear-time Minimum Bayes Risk Decoding with Reference Aggregation
This work addresses efficiency for practitioners using MBR decoding in machine translation, though it is incremental as it builds on existing methods.
The paper tackles the high computational cost of Minimum Bayes Risk (MBR) decoding in text generation by approximating pairwise utility scores with aggregated reference representations, reducing complexity from O(n^2) to O(n) while preserving most quality gains.
Minimum Bayes Risk (MBR) decoding is a text generation technique that has been shown to improve the quality of machine translations, but is expensive, even if a sampling-based approximation is used. Besides requiring a large number of sampled sequences, it requires the pairwise calculation of a utility metric, which has quadratic complexity. In this paper, we propose to approximate pairwise metric scores with scores calculated against aggregated reference representations. This changes the complexity of utility estimation from $O(n^2)$ to $O(n)$, while empirically preserving most of the quality gains of MBR decoding. We release our source code at https://github.com/ZurichNLP/mbr