Instructions on how to run the experiments from (Schlangen et al.,
2009)


1. Dependencies
   SRILM
   - install SRILM ( http://www.speech.sri.com/projects/srilm/ )
     (we ran our experiments with version 1.5.7; later versions
      are probably ok as well)
   - point environment variable SRILM_HOME to directory with SRILM
     binaries
   Statistics::Lite perl module

2. in directory scripts/ , run

   - for n-best:
      ./crosseval.pl -data=utt.mul.numnorm -ngram=2 -silcount -baseres=DistUpdate -decider=nbest

   - for rnd-nb:
      ./crosseval.pl -data=utt.mul.numnorm -ngram=2 -silcount -baseres=DistUpdate -decider=randnbest

   - for adapt:
      ./crosseval.pl -data=utt.mul.numnorm -ngram=2 -silcount -baseres=DistUpdate -decider=adpt0.1

   - for max:
      ./crosseval.pl -data=utt.mul.numnorm -ngram=2 -silcount -baseres=DistUpdate -decider=max

   - for random:
      ./crosseval.pl -data=utt.mul.numnorm -ngram=2 -silcount -baseres=Random

   You will find the result summaries in scripts/Results (with
   filenames indicting the parameters), the detailed
   logs in scripts/Logs.