RandLM now supports language models that are served on multiple machines. This means that language models can be very large, they now have a zero-time start up when used in Moses and they can be shared across multiple decoders. As they say in the trade, not bad.
http://sourceforge.net/projects/randlm/ Note that batching in the decoder (ie changing the search strategy) has not been implemented yet. Significant effort has gone into making the LM itself time and space efficient, Miles and Oliver -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
