Best-First Beam Search
This work provides a significant speed improvement for beam search in NLP tasks, benefiting researchers and practitioners who rely on efficient decoding algorithms, though it is incremental as it builds on existing beam search methods.
The paper tackles the inefficiency of standard beam search in NLP decoding by introducing a faster variant that prunes hypotheses early under monotonic scoring assumptions, achieving up to 10x speedup in practice while maintaining similar performance.
Decoding for many NLP tasks requires an effective heuristic algorithm for approximating exact search since the problem of searching the full output space is often intractable, or impractical in many settings. The default algorithm for this job is beam search -- a pruned version of breadth-first search. Quite surprisingly, beam search often returns better results than exact inference due to beneficial search bias for NLP tasks. In this work, we show that the standard implementation of beam search can be made up to 10x faster in practice. Our method assumes that the scoring function is monotonic in the sequence length, which allows us to safely prune hypotheses that cannot be in the final set of hypotheses early on. We devise effective monotonic approximations to popular nonmonontic scoring functions, including length normalization and mutual information decoding. Lastly, we propose a memory-reduced variant of Best-First Beam Search, which has a similar beneficial search bias in terms of downstream performance, but runs in a fraction of the time.