Abstract

We present a freely available open-source toolkit for
training recurrent neural network based language models. It can
be easily used to improve existing speech recognition and machine
translation systems. Also, it can be used as a baseline for future
research of advanced language modeling techniques. In the paper,
we discuss optimal parameter selection and different modes of
functionality. The toolkit, example scripts and basic setups are
freely available at http://rnnlm.sourceforge.net/.

‚Äč