Sampling-Based Approximations to Minimum Bayes Risk Decoding for Neural Machine Translation
This work addresses a key decoding problem in neural machine translation that affects translation quality, offering a method to improve it, though it is incremental as it builds on existing MBR approximations.
The paper tackles the problem of inadequate translations selected by beam search in neural machine translation, known as the beam search curse, by analyzing and designing approximations to minimum Bayes risk decoding that avoid this issue and allow for larger hypothesis spaces, showing benefits across three language pairs including English-German, English-Romanian, and English-Nepali.
In NMT we search for the mode of the model distribution to form predictions. The mode and other high-probability translations found by beam search have been shown to often be inadequate in a number of ways. This prevents improving translation quality through better search, as these idiosyncratic translations end up selected by the decoding algorithm, a problem known as the beam search curse. Recently, an approximation to minimum Bayes risk (MBR) decoding has been proposed as an alternative decision rule that would likely not suffer from the same problems. We analyse this approximation and establish that it has no equivalent to the beam search curse. We then design approximations that decouple the cost of exploration from the cost of robust estimation of expected utility. This allows for much larger hypothesis spaces, which we show to be beneficial. We also show that mode-seeking strategies can aid in constructing compact sets of promising hypotheses and that MBR is effective in identifying good translations in them. We conduct experiments on three language pairs varying in amounts of resources available: English into and from German, Romanian, and Nepali.