Аннотация:This paper presents novel approaches to phrase alignment for example-based machine translation (EBMT). We use matching of delimiters instead of word matching while determining fragment borders. Than we construct the critical path within two-dimensional bilingual space using dynamic programming. We follow a monotonic machine translation approach, for which we develop a flexible partial reordering with constraints. Experimental results show the significant improvement of translation quality.
Keywords: parallel sentences, translation, critical path, local inversion