Coursera Notes

Week 1

It introduced Language Model problems, mentioned using "perplexity" to evaluate the LM.

It showed the n-gram model and how the probability get calculated using linear interpolation for Trigram Model.

It mentioned applying the discounting method to Katz back-off model, a generative n-gram language model that estimates the conditional probability of a word given its history in the n-gram.

Week 2

Tagging Problem
From the training set, induce a function/algorithm that maps new sentences to their tag sequences.

Generative models for Supervised Learning
Bayes' rule to produce the conditional distribution
Discriminative Model:
Generative Model:

Hidden Markov Model (HNM) Taggers
basic definitions

parameter estimation

The Viterbi Algorithm: find the most likely hidden state sequence

It brought up the tagging problems and demonstrated how to apply Hidden Markov Model to find the most probable hidden state sequence(tags).

Week 3

Parsing Problem
work in formal syntax goes back to Chomsky's PhD thesis (synatictic sturctures) in 1957

context-free grammars
Using context free grammar could have ambiguity.