Improving MT Word Alignment Using Aligned Multi-Stage Parses
Adam Meyers, Michiko Kosaka, Shasha Liao, Nianwen Xue · Meeting of the Association for Computational Linguistics · 2011
We use hand-coded rules and graph-aligned logical dependencies to reorder English text towards Chinese word order. We obtain a 1.5% higher F-score for Giza++ compared to running with unprocessed text. We describe this research and its implications for SMT.