Accurate Online Posterior Alignments for Principled Lexically-Constrained Decoding
This work addresses the challenge of accurate online alignments for lexically constrained decoding in machine translation, which is crucial for applications like user-defined dictionary injection, but it is incremental as it builds on existing constrained beam-search algorithms.
The paper tackles the problem of online alignment in machine translation, which aligns target words to source words during partial decoding, by proposing a novel posterior alignment technique that jointly considers alignment and token probabilities. The method achieves consistent reductions in alignment error rates across five language pairs and significant BLEU improvements around constrained positions in seven lexically constrained translation tasks.
Online alignment in machine translation refers to the task of aligning a target word to a source word when the target sequence has only been partially decoded. Good online alignments facilitate important applications such as lexically constrained translation where user-defined dictionaries are used to inject lexical constraints into the translation model. We propose a novel posterior alignment technique that is truly online in its execution and superior in terms of alignment error rates compared to existing methods. Our proposed inference technique jointly considers alignment and token probabilities in a principled manner and can be seamlessly integrated within existing constrained beam-search decoding algorithms. On five language pairs, including two distant language pairs, we achieve consistent drop in alignment error rates. When deployed on seven lexically constrained translation tasks, we achieve significant improvements in BLEU specifically around the constrained positions.