Given a corpus of quotations, generate a new quotation using Language Model techniques

1. Final Report: CS 6700 Final Report.pdf
2. Presentation Link: CS 6700 Final Presentation.pdf
3. Proposal Link: CS 6700 Project Proposal.pdf

Tutorial: http://www.wildml.com/2015/09/recurrent-neural-networks-tutorial-part-1-introduction-to-rnns/ CUDA GPU on AWS instructions: https://github.com/dennybritz/rnn-tutorial-rnnlm Andrej Karpathy's RNN as LM: http://karpathy.github.io/2015/05/21/rnn-effectiveness/

• Python, pip v 2.7
• virtualenv

# Create a new virtual environment 
virtualenv venv
source venv/bin/activate

# Install requirements
pip install -r requirements.txt

# If Theano is giving you trouble (upgraded Mac Systems)
pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git

• Use simplified Kneser-Ney smoothing for bigram and trigrams (literature suggests it is superior to +1 laplace smoothing)
  • Subtract 0.75 from all n-grams which appear 2 or more times
  • Subtract 0.5 from n-grams which appear once
  • Add 0.000027 to unseen n-grams
  • https://www.researchgate.net/profile/Frankie_James2/publication/255479295_Modified_Kneser-Ney_Smoothing_of_n-gram_Models/links/54d156750cf28959aa7adc08.pdf
  • http://ieeexplore.ieee.org/abstract/document/4244538/?part=1

• Donald Trump: https://goo.gl/forms/yTv29gNEOguivH0B3
• Cicero: https://goo.gl/forms/eYgZmQsJMMdfATjw1

• http://www.wildml.com/2015/09/recurrent-neural-networks-tutorial-part-1-introduction-to-rnns/

• N/A

• http://www.wildml.com/2015/10/recurrent-neural-network-tutorial-part-4-implementing-a-grulstm-rnn-with-python-and-theano/