We discuss the relevance of k-best parsing to recent applications in natural language processing, and develop efficient algorithms for k-best trees in the framework of hypergraph parsing. To demonstrate the efficiency, scalability and accuracy of these algorithms, we present experiments on Bikel’s implementation of Collins ’ lexicalized PCFG model, and on Chiang’s CFG-based decoder for hierarchical phrase-based translation. We show in particular how the improved output of our algorithms has the potential to improve results from parse reranking systems and other applications. 1

A good decoding algorithm is critical

A good decoding algorithm is critical to the success of any statistical machine translation system. The decoder's job is to find the translation that is most likely according to set of previously learned parameters (and a formula for combining them). Since the space of possible translations is extremely large, typical decoding algorithms are only able to examine a portion of it, thus risking to miss good solutions. Unfortunately, examining more of the space leads to unacceptably slow decodings. In this

Introduction Crises and disasters frequently attract international attention to regions of the world that have previously been largely ignored by the international community. While it is possible to stock up on emergency relief supplies and, for the worst case, weapons, regardless of where exactly they are eventually going to be used, this cannot be done with multilingual information processing technology. This technology will often have to be developed after the fact in a quick response to the given situation. Multilingual data resources for statistical approaches, such as parallel corpora, may not always be available. In the fall of 2000, we decided to put the current state of the art to the test with respect to the rapid construction of a machine translation system from scratch. Within one month, we would # hire translators; # translate as much text as possible; and # train a statistical MT system on the data thus created. The languag

Abstract not found

We report on our experience with building a statistical MT system from scratch, including the creation of a small parallel TamilEnglish corpus, and the results of a taskbased pilot evaluation of statistical MT systems trained on sets of ca. 1300 and ca.