Title :
SyMGiza++: A tool for parallel computation of symmetrized word alignment models
Author :
Junczys-Dowmunt, M. ; Szal, A.
Author_Institution :
Fac. of Math. & Comput. Sci., Adam Mickiewicz Univ., Poznan, Poland
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 each iteration 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 MGiza++´s capability of parallel computations.
Keywords :
multiprocessing systems; parallel algorithms; software tools; SyMGiza++; multiprocessor system; parallel computation; symmetric word alignment model; symmetrized word alignment model; training algorithm; Computational modeling; Data models; Hidden Markov models; Mathematical model; Solid modeling; Training; Viterbi algorithm;
Conference_Titel :
Computer Science and Information Technology (IMCSIT), Proceedings of the 2010 International Multiconference on
Conference_Location :
Wisla
Print_ISBN :
978-1-4244-6432-6
DOI :
10.1109/IMCSIT.2010.5679866