Pushdown Automata Outperform Traditional Models in Statistical Machine Translation

Pushdown automata in statistical machine translation (2014)

A 2014 study in Computational Linguistics demonstrates that incorporating pushdown automata into statistical machine translation significantly improves translation quality. By modeling hierarchical structures beyond the capacity of regular grammars, the approach captures long-distance dependencies and recursive patterns. Experiments on multiple language pairs show consistent gains over phrase-based and hierarchical baselines, suggesting a promising direction for syntax-aware translation models.