##########
# nhpylm #
##########

python bindings for a c++ based implementation of the Nested Hierarchical Pitman-Yor Language model


###########
# Contact #
###########

In case of questions, suggestions, problems etc. please send an email or check the disussion group.

Oliver Walter:
walter@nt.uni-paderborn.de

Jahn Heymann:
heymann@nt.uni-paderborn.de

Discussion group:
email: latticewordsegmentation@googlegroups.com
google groups: https://groups.google.com/d/forum/latticewordsegmentation


##############
# References #
##############

Teh, Yee Whye. "A hierarchical Bayesian language model based on Pitman-Yor processes."
Proceedings of the 21st International Conference on Computational Linguistics
and the 44th annual meeting of the Association for Computational Linguistics.
Association for Computational Linguistics, 2006.


Mochihashi, Daichi, Takeshi Yamada, and Naonori Ueda. "Bayesian unsupervised word segmentation
with nested Pitman-Yor language modeling."
Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL
and the 4th International Joint Conference on Natural Language Processing of the AFNLP:
Volume 1-Volume 1. Association for Computational Linguistics, 2009.


###############
# Instalation #
###############

git clone [...] nhpylm
cd nhpylm
pip install --user -e .


################
# Requirements #
################

kaldi: to run the examples, the openfst command line tools installed by kaldi will be needed.
Prior to running the examples, set the environment varaible KALDI_ROOT to point to you kaldi installation.
(see nhpylm/kaldi.py)


############
# Examples #
############

Example output can be found in examples/lattice_playground
All possible segmentations of an example sentence: examples/lattice_playground/wcl-l-e-p/I_loop_L.fst.pdf

Example using PHI compositions: examples/Word_Char_Lexicon-LM-Example-Phi.ipynb
Example using PHI compositions and acoustig model: examples/Word_Char_Lexicon-LM-Example-Phi-Monophone.ipynb
Example using EPS fallback (standard decoding chain): examples/Word_Char_Lexicon-LM-Example.ipynb

PHI composition with example from WSJ database: examples/Word_Char_Lexicon-LM-WSJCAM0-Phi.ipynb
EPS fallback with example from WSJ database: examples/Word_Char_Lexicon-LM-WSJCAM0.ipynb
Decoding of test sentences with both, PHI composition and EPS fallback: examples/Word_Char_Lexicon-LM-WSJCAM0-Decode.ipynb

Create (L o G) for WSJCAM0 training data: examples/WSJCAM0_HPYLM.ipynb
Create (L o G) for WSJ LM training data: examples/WSJ_HPYLM.ipynb

Example using PHI compositions and character only language model: examples/Char-LM-Example-Phi.ipynb