Multilingual Alignments by Monolingual String Differences

We propose a method to obtain subsentential alignments from several languages simultaneously. The method handles several languages at once, and avoids the complexity explosion due to the usual pair-bypair processing. It can be used for different units (characters, morphemes, words, chunks). An evaluation of word alignments with a trilingual machine translation corpus has been conducted. A comparison of the results with those obtained by state of the art alignment software is reported.

• Type: Conference
• Year: 2008
• Where: COLING
• Authors: Adrien Lardilleux, Yves Lepage
• Keywords: COLING 2008, Computational Linguistics, Subsentential Alignments, Trilingual Machine Translation, Usual Pair-bypair Processing