[moses-smt/mosesdecoder] 45a235: avoid speed loss when working with unpruned models...

[GitHub <no...@gi...>]

  Branch: refs/heads/matecat_multimodel
  Home:   https://github.com/moses-smt/mosesdecoder
  Commit: 45a2356273e6bf6b3e4d3903e3cd877ddc8d7e6c
      https://github.com/moses-smt/mosesdecoder/commit/45a2356273e6bf6b3e4d3903e3cd877ddc8d7e6c
  Author: Rico Sennrich <ric...@gm...>
  Date:   2012-10-31 (Wed, 31 Oct 2012)

  Changed paths:
    M moses/src/PhraseDictionaryMultiModel.cpp
    M moses/src/PhraseDictionaryMultiModel.h

  Log Message:
  -----------
  avoid speed loss when working with unpruned models by applying ttable-limit to each component model.
this more aggressive pruning may give slightly different results than before,
but overhead of online interpolation of *all* translation options is simply too big.