Companion page to "Predictive Incremental Parsing Helps Language Modeling"

This page contains reference material and will be expanded to allow replication of our findings in our COLING 2016 paper. If you have questions about our work, contact Arne Köhn or Timo Baumann.

Software

The parser we used is described in "Incremental Predictive Parsing with TurboParser", the code is available on its companion page. We also used the same models for English.

For the N-gram modeling reported in Section 5.1, we used SRILM. We happily provide you with the scripts used to create the sub-models.

We adapted faster-rnnlm to carry out the experiments in Section 5.2 and 5.3. We happily provide you with the patch.

Data

All experiments were carried out on the billion word corpus.

We happily provide you with the parsed version of the corpus. (We'll have to check whether we can make it available online, given its size. Mailing you a flash drive will probably be the better choice.)