Hidden-Markov-Models

Michael Collins's NLP Coursera Course - Lab #1

Goal: Accurately generate part of speech tags using a trigram Hidden Markov Model. In this application, I distinguish between 'I-GENE' and normal words in biological text.

Tasks:

1. Figure out which words in the training data occur < 5 times, and can be used to estimate counts for rare words to smooth the probabilities. Replace those words in the counts file with 'RARE'. See replace_rare.py
2. Compile the unigram, bigram and trigram probabilities. See emission_probs.py
3. Implement the Vitterbi Algorithm to generated the most likely tags given the training data. Also see emission_probs.py
4. Classify rare types based on capitalization, digits, etc. SEe adv_replace_rare.py