Abstract | ||
---|---|---|
SyMGiza++ -- a tool that computes symmetric word alignment models with the capability to take advantage of multi-processor systems -- is presented. A series of fairly simple modifications to the original IBM/Giza++ word alignment models allows to update the symmetrized models between chosen iterations of the original training algorithms. We achieve a relative alignment quality improvement of more than 17% compared to Giza++ and MGiza++ on the standard Canadian Hansards task, while maintaining the speed improvements provided by the capability of parallel computations of MGiza++. Furthermore, the alignment models are evaluated in the context of phrase-based statistical machine translation, where a consistent improvement measured in BLEU scores can be observed when SyMGiza++ is used instead of Giza++ or MGiza++. |
Year | DOI | Venue |
---|---|---|
2011 | 10.1007/978-3-642-25261-7_30 | international workshop on security |
Keywords | Field | DocType |
chosen iteration,symmetric word alignment model,statistical machine translation,consistent improvement,original ibm,alignment model,speed improvement,symmetrized word alignment model,original training algorithm,word alignment model,bleu score,relative alignment quality improvement | IBM,Computer science,Machine translation,Phrase,Algorithm,Speech recognition,Computation | Conference |
Volume | ISSN | Citations |
7053 | 0302-9743 | 27 |
PageRank | References | Authors |
1.30 | 12 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Marcin Junczys-Dowmunt | 1 | 312 | 24.24 |
Arkadiusz Sza$#322/ | 2 | 27 | 1.30 |